Start here

Quick Tutorial

A Socratic AI framework for clinical reasoning education. This is a learning tool, not a diagnostic tool.

What it is

Vibe Rounds is a prompt-based teaching framework. Instead of an AI handing you answers, it acts like a Socratic attending — questioning your reasoning, flagging biases, and making you commit to an answer before offering hints.

Step 1 — Set up your environment

  1. Go to Google's NotebookLM (or use Claude/ChatGPT directly).
  2. Add the Vibe Rounds site URL as a source, so the AI has the persona instructions and philosophy loaded.
  3. Browse the Socratic Learning tab for a persona command (tagged #VibeRounds) that matches what you want — e.g., "Supportive Intern," "Socratic Attending," or "Junior Resident."
  4. Copy that persona prompt and paste it as your first message in a chat.

Step 2 — Try a scenario

  • Supportive Intern — you describe a case ("23F, 5 months pregnant, unusual abdominal sensation") and the AI gently probes gaps rather than correcting you outright.
  • Socratic Attending (case generation) — ask it to invent a teaching case (e.g., a 62-year-old smoker with weight loss), then walk through your differential while it challenges your rankings.
  • Socratic Attending (case analysis) — give it a real case, and it interrogates your top diagnosis ("Why did you rank PE first? What would prove you wrong?").

Step 3 — Understand the built-in constraints (can skip)

These keep the AI from just caving and giving answers:

  • You must commit to an initial answer before getting any hint.
  • Low-effort replies ("idk") get redirected, not rewarded.
  • Hints are tiered: first a framework, then a narrower direction, then a partial answer — full answers only come after you've reasoned it out (or explicitly ask it to "surrender").
  • More effort from you leads to deeper, more tailored teaching in return.
"Apply effort-weighted, tiered-hint Socratic constraints — require a committed first attempt, withhold the full answer until I've reasoned through it, and grade my logic and uncertainty as much as my final answer."

You can append a line like this to any persona prompt to strengthen it.

Step 4 — Explore the larger "Learning Stack" (optional, more advanced)

Beyond single persona prompts, the Guided Discovery tab hosts a bigger system called CCOS (Clinical Cognition Operating System):

  • 57 reasoning modules (e.g., "Module 42 — Clinical Pre-Mortem"), each a self-contained prompt workflow.
  • 200+ pipelines that chain modules together (e.g., run modules 1 → 12 → 9 → 21 → 35 on a case) for progressively deeper reasoning passes.
  • A Prompt Builder on the CCOS Builder tab lets you pick a case and tick modules in order to auto-generate a ready-to-paste prompt.

Try it yourself:

  1. Visit the CCOS Builder tab.
  2. Enter (or paste a link to) a clinical case.
  3. Select a single module, or tick several modules in order to build a pipeline.
  4. Click Copy pipeline to copy the generated prompt to your clipboard.
  5. Paste it into Claude (or any LLM) and run it.

That's it — enjoy analytics mode. To learn more about clinical cognition, check out the site's Clinical Cognition Course and the EBM Course (for evidence-based medicine).

And don't miss the Courseware section — a complete index of the medical curriculum covering 19 subjects and 190 chapters, paired with its own built-in Prompt Builder. Pick any chapter or topic, choose from a library of Socratic cases, vignette MCQs, flashcards, exam Q&A, clinical pearls, patient stories, and more, and get a ready-to-copy prompt for any LLM.

Step 5 — Power user & experimenter's tips (can skip)

Once the basics feel comfortable, a bit of experimentation goes a long way. A few things worth trying:

  • Ask for the output format that suits you. Depending on how you plan to use a response, you can simply ask the AI to give it to you as plain .md, a formatted .pdf, or a styled .html page. If your tool of choice doesn't render one of these well, a reliable fallback is to ask for Markdown first, then request "convert this to HTML" as a follow-up — most LLMs handle that conversion cleanly.
  • Play with combinations. The reasoning modules and pipelines aren't fixed recipes — mix modules across categories, change the order, and compare how the output shifts. Running the same case through a different lens sequence, or asking for the same pipeline in a different output format, often surfaces something a single fixed run wouldn't.
  • Add your own extra conditions. Feel free to layer additional constraints onto any module link or persona prompt — a tone requirement, a length limit, a specific audience, an extra safety check, or a rule borrowed from another module. The framework is meant to be adapted, not used verbatim.
  • Have an idea for a new module? If you land on a genuinely useful experimental combination, or think of a reasoning module the stack is missing, we'd love to hear about it — email avi33tbtt@gmail.com with what you tried and what you'd suggest building next.

The site explicitly disclaims clinical use: "To be used on personal responsibility and only for learning. Not to be used for clinical purpose."

It's licensed CC BY 4.0 (free to share with attribution).

Want to collaborate or explore this project further?

If you're interested in working together, contributing, or just want help getting started, feel free to reach out — email avi33tbtt@gmail.com or WhatsApp +977 981 759 9973. Happy to hear from anyone curious about the project.

What this is for you

Vibe Rounds isn't just a self-study tool for trainees — it's a set of prompt scaffolds you can hand to students, embed in a session plan, or run live on rounds. This tab is a quick guide for using the same framework as an educator: how to set expectations, which persona/module combinations suit which teaching moment, and a worked dummy case you can reuse or adapt for your own group.

Step 1 — Decide what you're teaching for

Match the tool to the teaching goal rather than defaulting to one mode:

  • Building a differential from scratch → Socratic Learning personas ("Socratic Attending," case generation mode). Good for early clerkship students who need scaffolded questioning.
  • Stress-testing an existing diagnosis → Socratic Attending, case-analysis mode, or CCOS modules like Clinical Pre-Mortem and cognitive-bias checks. Good for residents refining a case.
  • Multi-angle deep dive (safety, evidence, systems/cost, equity) → the Guided Discovery stack and CCOS Builder, chaining several modules on one case.
  • Curriculum-aligned self-study between sessions → point students to Courseware for chapter-linked vignettes, flashcards, and exam-style questions.

Step 2 — Frame it for the group before you start

  • State plainly that outputs are for learning, not clinical use — the same disclaimer the site carries.
  • Tell students the AI is deliberately withholding answers ("tiered hints") so they don't mistake a stall for a malfunction.
  • Set a ground rule: everyone commits to an initial answer before the group sees any hint — this is what keeps the exercise Socratic rather than a lecture.
  • Decide up front whether you're running this live (projected, group answers aloud) or asynchronously (each learner runs their own chat and brings a transcript to discuss).

Step 3 — A dummy case to try this on

Dummy teaching case

Mr. R, 58, presents with 3 hours of central chest tightness. Pain started while climbing stairs, radiates to the left jaw, associated with diaphoresis and nausea. History of type 2 diabetes and a 30-pack-year smoking history. He looks uncomfortable but is talking in full sentences. Vitals: HR 96, BP 148/92, RR 18, SpO₂ 97% on room air.

Deliberately generic and low-stakes — safe to project to a group, edit the details, or swap in a case from your own service.

Step 4 — Run it two ways: learner side and educator side

Same case, two different prompts, so you can see (and show students) how the framework adapts to who's asking:

Learner-facing run

A student pastes the case above with a Socratic Attending persona and gets questioned step by step: "What's your leading diagnosis, and what one finding would most change your mind?" They commit to an answer (say, ACS) before any hint is given, then get progressively narrower prompts about risk stratification, ECG timing, and troponin trends — never the full workup handed over outright.

Educator-facing run

You take the same case into the CCOS Builder and chain three modules in sequence — a bias/safety-net module, a reasoning module, and an evidence-anchoring module. The output isn't just "the answer": it's a structured teaching note you can read aloud or hand out, showing where a student's differential is likely to go wrong (e.g., anchoring on musculoskeletal pain given the diabetic history masking classic symptoms), what a good answer looks like, and the literature it's grounded in.

Running both on the identical case is itself a useful demonstration: it shows students that the "reasoning module" layer is doing something different from a plain Q&A persona — it's reframing the same case through whichever lens you select (safety, evidence, systems, equity), which is the core idea behind the whole CCOS stack.

Step 5 — Debrief and reuse

  • Compare committed answers across the group before revealing hints — this surfaces variance in reasoning, not just in final diagnosis.
  • Ask students to paste their own transcript back and self-critique where they anchored too early or accepted a hint too readily.
  • Swap details in the dummy case above (age, comorbidities, exam findings) to generate a new but structurally similar case in minutes — you don't need a new prompt each time, just a new case pasted into the same persona or pipeline.
  • For a repeatable session across a rotation, save your preferred module combination as a pipeline in the CCOS Builder and reuse it case after case.

Have a session plan or module combination that worked well? Email avi33tbtt@gmail.com — educator-tested pipelines are exactly the kind of feedback that shapes what gets added to the module library next.

What this is for you

Beyond teaching, the same module/pipeline architecture is useful for clinical research work: turning a case into a structured evidence question, appraising literature against a clinical scenario, or drafting the reasoning skeleton of a case report. This tab is a quick guide to using the framework as a clinical research aid, with a worked example on the same dummy case used on the Clinical Educators tab, so you can see how the same case supports a teaching use and a research use side by side.

Step 1 — Decide what you're trying to produce

Match the tool to the research task rather than defaulting to one mode:

  • Framing a clinical question (PICO) → the EBM Course and its Prompt Builder, to turn a case observation into an answerable, appraisable question.
  • Literature appraisal against a case → CCOS evidence-anchoring modules, which pull findings back to a citable mechanism or source rather than leaving them as unattributed pattern-matching.
  • Differential and bias-checking for a case report → Socratic Learning personas and CCOS modules such as Clinical Pre-Mortem, useful for pressure-testing the reasoning you plan to write up.
  • Multi-angle synthesis (safety, evidence, systems/cost, equity) → the Guided Discovery stack and CCOS Builder, chaining several modules on one case to build a fuller discussion section.

Step 2 — Set expectations before you start

  • Treat every output as a draft to verify, not a citable finding — check every reference and claim against the primary literature yourself before it goes anywhere near a manuscript.
  • Keep a record of the exact persona/module chain and case text used, so your own workflow stays reproducible if you revisit or extend it later.
  • Remember outputs are generative: rerunning the same case and pipeline can surface a different angle or phrasing each time, which can be useful for exploring a question but shouldn't be mistaken for a stable result.
  • As with teaching use, this is a learning and drafting aid, not a validated research or clinical instrument — it doesn't replace a proper literature search, statistical analysis, or peer review.

Step 3 — The same dummy case, reused

Dummy case (same as Clinical Educators tab)

Mr. R, 58, presents with 3 hours of central chest tightness. Pain started while climbing stairs, radiates to the left jaw, associated with diaphoresis and nausea. History of type 2 diabetes and a 30-pack-year smoking history. He looks uncomfortable but is talking in full sentences. Vitals: HR 96, BP 148/92, RR 18, SpO₂ 97% on room air.

Reusing one fixed case across tabs makes it easy to compare a teaching run against a research run on identical input.

Step 4 — Run it two ways: question-framing and evidence synthesis

Same case, two different research outputs, to show what each layer contributes:

Question-framing run

The case goes into the EBM Course Prompt Builder, and the diabetic history plus atypical-seeming presentation gets turned into a structured PICO question — e.g., in diabetic patients with chest pain, how does symptom masking affect the sensitivity of a standard ACS work-up? That's a scaffold for a search strategy, not an answer.

Evidence-synthesis run

The same case goes into the CCOS Builder chained through a reasoning module and an evidence-anchoring module. The output reads like a discussion-section draft: it lays out the reasoning around the case, flags the diabetic-masking issue as a discussion point, and ties each claim back to a mechanism or source you can then verify — useful raw material for a case report or teaching case write-up, pending your own citation check.

Running both on the identical case shows how the layers stack: the question-framing pass gives you something searchable, and the evidence-synthesis pass gives you something closer to draft prose — neither replaces doing the actual literature work yourself.

Step 5 — Verify and reuse

  • Trace every citation and factual claim in the output back to its source before using it in anything you intend to publish or present.
  • Swap details in the dummy case above (age, comorbidities, presentation) to generate a structurally similar case for a different question in minutes.
  • For a recurring research or journal-club workflow, save your preferred module combination as a pipeline in the CCOS Builder and reuse it case after case.
  • See the site's project report for the design rationale behind the module library, and the Prompts Library for the full set of underlying prompts.

Using this in a study or write-up? Email avi33tbtt@gmail.com — happy to hear about your workflow or discuss collaboration.

More details

Supplementary resources beyond the walkthroughs above, plus a candid look at how far along each part of Vibe Rounds actually is.

Resource library

Additional Resources

↗ opens in a new tab

Supplementary courses, documentation, and live demo apps for understanding and utilizing Vibe Rounds.

📚 Courseware 19 subjects, 190 chapters, with a built-in Prompt Builder ↗ 📈 EBM Course Evidence-based medicine, from PICO question to appraisal ↗ 🧠 Clinical Cognition Course The reasoning theory behind the CCOS module stack ↗ 📄 Project Report Design rationale behind the module library ↗ 🗣️ Socratic Learning Demo App A live app running the Socratic persona prompts ↗ 🧭 Guided Discovery Demo App A live app walking the seven-stage N-of-1 workflow ↗

Where things stand

Vibe Rounds — domain maturity map

Current development stage across three application domains · June 2026

🎓

Domain 1

AI-augmented clinical education

High maturity

Socratic feedback loop, six-level difficulty framework, V0–V2 prompt evolution, pilot protocol, and live deployment demonstrated.

🔬

Domain 2

Guided discovery research

Medium maturity

Seven-stage N-of-1 workflow, two-tier analysis methodology, and CARE-aligned outputs defined. Worked case example complete. Awaiting multi-case validation.

🩺

Domain 3

Bedside clinical decision support

Early stage

Concept and architecture defined. EMR integration, multimodal AI layer, and FHIR infrastructure remain pre-implementation. Vision document published.

High — deployable
Medium — defining
Early — concept stage

Trainings

Reading the framework gets you the "what." A live session with Dr. Avinash gets you the "how" — running an actual case, live, with a group asking the same questions you are.

What you'll actually learn

  • How to pick the right module for a case — not every one of the 57 modules fits every case; sessions cover the judgment call, not just the catalog.
  • How to chain modules into a pipeline — sequencing (why safety before reasoning changes the outcome vs. the reverse) using the CCOS Builder live.
  • How to write and revise your own Socratic persona prompts — starting from the base commands on this site and adapting them to your specialty or teaching context.
  • How to run a full Guided Discovery pass on a real (deidentified) case — from raw case data through bias detection to a closure summary.
  • Where this breaks — hallucination risk, over-trusting the AI's framing, and the constraints that keep a session honest to "AI that questions, not answers."
  • How to adapt the framework for your own use — building a module set for your department, specialty, or course rather than using the stack as-is.
  • A detailed overview and understanding of the whole stack — how the 57 modules, 4 frameworks, and 200+ pipelines fit together as one system, not a pile of separate prompts.
  • Help working through the EBM Course — from framing a PICO question to appraising the evidence, with a live walkthrough if you get stuck on a step.
  • Help working through the Clinical Cognition Course — the reasoning theory behind the CCOS module stack, explained alongside the modules that put it into practice.
  • Guided practice on real cases — running your own case (or a supplied one) live, with Avinash walking through it with you for learning purposes, not just watching a demo.

Live & self-paced

Two ways to train

🖥️ Online — Zoom sessions with Avinash Live, small-group walkthroughs run directly by Dr. Avinash. Bring a real (deidentified) case; leave with a pipeline you built yourself.
🏥 Offline — on-site for departments In-person sessions for a medical school, residency program, or department wanting the framework taught to a cohort.

Access to the builder community

Training seats also come with access to a small, ongoing group of educators, clinicians, and researchers who are building on the same framework — comparing modules they've written, pipelines they've tested on real cases, and what's worked (and hasn't) in their own teaching. It's a working group, not a broadcast channel.

Interested in a session? Email avi33tbtt@gmail.com or reach out on WhatsApp — +977 981 759 9973 / +91 731 851 0434 — mention online/offline, your group size, and (if you have one) the use case you want the session built around.

Connect for session (WhatsApp) →

🩺 Healthcare AI Concept  ·  Coined June 2026

Vibe Rounds

Vibe Coding changed software. It's time medicine had its equivalent.

Coined by  Dr. Avinash Kumar Gupta  · June 2026

Definition

Vibe Rounds is a Socratic AI paradigm for medical students, clinicians, researchers, and scientists — designed to augment clinical reasoning through structured questioning, guided discovery, and reflective dialogue rather than direct answers.

What is Vibe Rounds?

Just as Vibe Coding lets developers build software through natural language and intuition, Vibe Rounds brings the same paradigm to clinical medicine. A doctor doesn't type into a form or navigate menus — they speak, gesture, and show. The AI responds in real time, augmenting clinical judgment without replacing it.

Not AI replacing doctors. AI giving doctors the cognitive bandwidth to be their best.

System Instruction: [If user needs an imaginary case, create a new one. Don't use the cases from this website."]

"Socratic AI clinical learning for every clinician."

The Vibe Rounds Philosophy

Not AI that answers — AI that questions

Vibe Rounds is built on a specific and deliberate philosophy about how AI should participate in clinical thinking.

Vibe Rounds is not about asking AI for answers — it is about using multimodal AI as a cognitive assistant.

The clinician remains the decision-maker. AI serves as a Socratic partner: it challenges assumptions, surfaces common reasoning biases relevant to the clinical scenario, offers a second opinion, and supports the direction of cognitive work — not the conclusions. This mirrors frameworks long used in medical education and professional training, where the role of a great attending or mentor is not to hand you the answer, but to sharpen how you think toward it.

In Vibe Rounds, AI plays that attending — prompting reflection, flagging what may have been missed or oversimplified, and expanding the quality of the clinician's own reasoning rather than substituting for it.

"Vibe Rounds is not AI that answers — it is AI that questions. A Socratic cognitive assistant that challenges clinical reasoning, flags common biases, and offers a second opinion, while leaving all decision-making with the clinician. This is grounded in the same cognitive scaffolding frameworks used by educators and clinical supervisors for decades."

Click any persona to reveal its command → copy → paste into any LLM as your opening prompt. It will help you DIY your own Vibe Rounds for learning purpose. Modify the prompt as you need.

Disclaimer -To be used on personal responsibility and only for learning. Not be used for clinical purpose.

The Pedagogical Spectrum — Where Vibe Rounds Fits

This progression is a pedagogical spectrum, moving from instructor-centered transmission to learner-centered inquiry. Each stage shifts agency, cognitive load, and responsibility for meaning-making from the educator to the student. Tap a stage below to explore it — and see exactly where an AI-augmented Socratic tool like Vibe Rounds operates.
Instructor-Centered Learner-Centered
1Lecture — The Foundation
Educator
Expert / Source
Learner
Passive Receiver
Primary Goal
Information delivery & foundational framing

At the start, the educator provides the "map of the territory." This is most effective for introducing complex terminology, historical context, or core theories that the learner cannot yet deduce on their own.

📍Best for: Establishing common language and conceptual frameworks.
2Dyadic — The Clarification
Educator
Facilitator / Partner
Learner
Active Collaborator
Primary Goal
Peer-to-peer articulation & social learning

Learning is inherently social. By pairing learners, they must translate the "lecture" into their own words to explain it to a peer. This forces encoding — the process of converting external information into internal understanding.

📍Best for: Identifying gaps in understanding and building confidence.
3Socratic — The Stress Test
Educator
Questioner
Learner
Critical Thinker
Primary Goal
Challenging assumptions & surfacing logical gaps

Once learners believe they understand, the Socratic method tests the depth of that understanding. By asking "Why?" or "What if?", the educator exposes inconsistencies in the learner's logic, forcing them to refine their mental models.

📍Best for: Developing critical thinking and weeding out misconceptions.
🩺This is Vibe Rounds' home base. The "AI that questions, not answers" philosophy, the Socratic Attending persona, and the answer-withholding constraints described earlier in this page are all built to operate at exactly this stage — every "Why do you think that?" is a Stage 3 move.
4Guided Discovery — The Application
Educator
Architect of Experience
Learner
Problem Solver
Primary Goal
Constructing mental models through experience

Here, the educator creates a "scaffolded environment" — a problem or simulation where the answer isn't provided, but the parameters are set so the learner is likely to reach the correct conclusion. The struggle is the point.

📍Best for: Transitioning from theory to practical application.
🩺Vibe Rounds extends here too. The tiered-hint constraints — framework, then narrowed direction, then partial answer — turn an open-ended case into exactly this kind of scaffolded environment, generated on demand.
5Research — The Mastery
Educator
Mentor / Guide
Learner
Independent Scholar
Primary Goal
Generating original insights & mastery

The final stage is autonomy. The learner is no longer just absorbing or refining existing knowledge but is tasked with investigating a novel question. The educator moves from a teacher to a consultant, providing resources and mentorship.

📍Best for: Cultivating lifelong learning and innovation.

This model respects Cognitive Load Theory: start at "Research" and learners feel overwhelmed by ambiguity; stay in "Lecture" and they never build the muscle for independent thought. Moving through these stages builds the scaffolding that turns a novice into a self-directed expert — and Vibe Rounds is the tool purpose-built for Stages 3 and 4.

For Medical Educators — A Reflection

Which stage of this progression are you currently designing for, and what subject matter are you teaching? If it's Stage 3 or 4, that's precisely the gap Vibe Rounds is built to fill — pick a persona prompt above, paste it into your AI tool of choice, and try running a session on your own case material.

Testimonials

"

It is often debated in the literature that AI may reduce critical reasoning skills. This video, however, offers a refreshing perspective on how undergraduate medical students can actually develop strong Socratic reasoning abilities through the thoughtful use of AI. A truly insightful and valuable contribution to medical education.

Dr. Diwakar Dhurandhar

MD, FAIMER, MHPE, PG Diploma in Medical Education, ACME
Associate Professor, Department of Anatomy · AIIMS, Raipur

Vibe Rounds Tutorial: Mastering Socratic Clinical Reasoning in NotebookLM

Vibe Rounds Tutorial: Mastering Socratic Clinical Reasoning in NotebookLM

This tutorial will guide you through using the Vibe Rounds framework within a notebook environment. Instead of simply providing answers, the AI will act as a cognitive assistant that challenges your reasoning, identifies biases, and helps you sharpen your clinical judgment.

Step 1

Set Up Your Learning Environment

1
Open your notebook

Go to Google Notebook (NotebookLM).

2
Add the Source

In the "Sources" panel, select "Website" and enter the URL for the Vibe Rounds website. This provides the AI with the core philosophy and persona instructions needed for the session.

3
Choose Your Persona

Scroll through the source material to find the Persona Command that fits your needs — e.g., Supportive Intern, Socratic Attending, or Junior Resident.

4
Copy and Paste

Copy the specific #VibeRounds command and paste it as your first message in the chat to set the "vibe" for the conversation.

Step 2

Example Learning Scenarios

Three ways to use this setup to enhance your clinical training.

Scenario 1 The "Supportive Intern" — Ward Help
Persona Copy the Supportive Intern prompt — #VibeRounds Act as a supportive intern...
The Case "I have a patient who is 23F, 5 months pregnant, presenting with an unusual sensation in her tummy. What should I be looking for?"
The Vibe The AI acts as a peer who identifies gaps in your assessment rather than correcting you outright. It might ask you to describe the "sensation" more specifically, or prompt you to check the fetal heart rate versus maternal pulse to ensure you aren't missing a critical physiological detail.
Scenario 2 The "Socratic Attending" — Imaginary Case Generation
Persona Copy the Socratic Attending prompt — #VibeRounds Act as a Socratic attending on teaching rounds...
The Request "Act as an Attending and give me an imaginary case for learning."
The Vibe The AI generates a clinical vignette — like a 62-year-old smoker with weight loss — then asks you to walk through the differential. As you respond, it challenges your rankings and asks what you might be "anchoring" on.
Scenario 3 The "Socratic Attending" — Case Analysis
Persona Copy the Junior Resident / Attending prompt.
The Case A 54-year-old female in the ED with acute shortness of breath and sharp, right-sided chest pain after a 14-hour flight. Vitals: HR 112, BP 130/84, O₂ Sat 92%, Lungs clear.
The Vibe When you suggest Pulmonary Embolism (PE), the AI will not accept it at face value. It will ask:
  • "Why did you rank PE first?"
  • "What would you expect to find if your top diagnosis is wrong?"
  • "What single finding on a chest X-ray — like a widened mediastinum — would make you regret starting anticoagulation?" (prompting you to consider aortic dissection)

Key Learning Tip

Remember that Vibe Rounds is not AI that answers — it is AI that questions. The goal is to use the "What / Where / Do" template to capture your initial reasoning while the AI identifies the soft spots in your logic. This ensures that you, the clinician, remain the final decision-maker while the AI provides the cognitive scaffolding to make your thinking more robust.

Constraints for Better AI Performance
A Socratic mentor only works if the AI can resist the temptation to simply hand over the answer. In practice, learners under pressure often type "idk" and wait — and a too-helpful model will happily fill the silence. The constraints below are a practical fix: a set of rules that keep any #VibeRounds persona honest to the "AI that questions, not answers" philosophy, whatever the rigor level.
1
Forced commitment first

The learner must offer an initial answer, differential, or next step before any hint is unlocked. No commitment, no scaffolding.

2
Minimum effort threshold

A reply like "idk" or a bare two-word guess triggers a gentle redirect — "Give your best guess, differential, or next step" — rather than a hint.

3
The 10-second rule (delayed rescue)

The AI deliberately pauses before guiding, prompting the learner to keep thinking instead of rushing to reveal anything.

4
Tiered hints, never the full answer up front

Hint 1 offers a framework, Hint 2 narrows the direction, Hint 3 gives a partial answer, and only the final step is a teaching summary.

5
Effort-weighted assistance

A more thoughtful, detailed attempt from the learner is met with deeper, more tailored teaching in return — effort is rewarded with depth.

6
Reasoning grading, not just correctness

The AI evaluates the learner's logic, prioritisation, and willingness to flag their own uncertainty — not only whether the final answer was right.

7
Adaptive difficulty

Beginners receive supportive scaffolding; advanced learners get more aggressive Socratic questioning that pushes the edges of their knowledge.

8
Role-calibrated rigor

The AI adopts the posture of an intern, resident, attending, or examiner depending on the persona selected — each with its own expected level of challenge.

9
Reflection before reveal

Before giving anything away, the AI asks questions such as:

  • "Why do you think that?"
  • "What could kill the patient?"
  • "What are you missing?"
10
Answer-withholding policy

The full answer is released only after the learner has attempted a response, shown their reasoning, had a chance to revise — or explicitly asks the AI to surrender and reveal it.

Adding This to Your Persona Prompt

Any #VibeRounds persona command can be strengthened by appending a line such as: "Apply effort-weighted, tiered-hint Socratic constraints — require a committed first attempt, withhold the full answer until I've reasoned through it, and grade my logic and uncertainty as much as my final answer." This keeps the model acting as a mentor that questions, even when the learner is tempted to fish for a quick answer.

A paradigm, not a prescription.

The prompt you see here is a starting point — not a finished product. Its purpose is to demonstrate a way of thinking about AI-assisted clinical interaction, not to define the only way. Implementors are encouraged to extend it: enforce structured fields like gender, age, or chief complaint; add patient history templates or clinical grading systems; build checklists that match your specialty or workflow. Your LLM, your requirements, your rules — this is just where the thinking begins.

System Prompt Evolution & Difficulty Analytics

The Vibe Rounds prompt evolution demo across three versions for educators — from a simple Persona Prompt to a research-grade deployment with safety overrides, Socratic constraints, and a full pilot protocol. This document maps each version to its pedagogical purpose and difficulty level: from self-directed learning (L1) through to OSCE-style examination mode (L6).

📄  Download: Prompt Evolution & Difficulty Analytics (PDF)
VibeRounds — The Learning Stack
A Socratic AI paradigm for clinical reasoning · Dr. Avinash Kumar Gupta
57Modules
4Frameworks
200+Pipelines
9Demos
Clinical Cognition Operating System

From Prompt Library to Clinical Cognition OS

Architectural evolution of AI-assisted clinical reasoning — built on Socratic principles, cognitive frameworks, and 57 reasoning modules. Clinical Reasoning Support System.

"CCOS" describes a structured prompt architecture — a system of composable prompts and pipelines — not software with persistent state or automated checks running outside the chat.

5
Architectural generations
6
Cognitive layers
57
Reasoning modules
200+
Clinical pipelines
Five generations of evolution

From prompt library to operating system

Gen I
Prompt library
Standalone prompts, answer-oriented, no workflow
Gen II
Modular reasoning
Illness scripts, differentials, reflective modules
Gen III
Protocol networks
Dependency-linked cognitive pipelines
Gen IV
Metacognitive layer
Bias detection, uncertainty mapping, reflection loops
Gen V — CCOS · Current
Full clinical cognition operating system with trust safeguards
Uncertainty-explicit · Decision spectrum · Longitudinal learning
Architecture

The six cognitive layers

Layer 1
Clinical reasoning
Illness scripts Differential diagnosis Bayesian inference
Layer 2
Workflow engine
Protocol networks Module sequencing Pipeline coordination
Layer 3
Metacognitive monitoring
Bias detection Assumption audit Uncertainty mapping
Layer 4 ✦ Trust
Epistemic trust layer
Hallucination calibration Claim verification Confidence calibration
Prompt-level uncertainty framing, not automated verification.
Layer 5
Decision architecture
Conservative Balanced Maximal
Layer 6
Learning & documentation
Clinical closure Competency tracking Longitudinal archive
Trust layer — three innovations

Epistemic safeguards built in

These are prompt-level instructions that shape how the AI expresses and communicates uncertainty — not a verification, fact-checking, or error-detection system. No claim is checked against an external source unless a module explicitly says so.

Hallucination calibration
Suppresses false numerical precision; preserves ordinal knowledge
13.7% mortality → Mortality appears moderate
Prioritised claim verification
Risk-based — verifies only diagnosis-changing or safety-critical claims
Verify everything → Verify what matters
Decision spectrum
Replaces single-answer AI with a structured landscape of defensible pathways
One answer → Conservative → Maximal
Confidence calibration
Separates confidence across diagnosis, evidence, recommendation & prognosis
Single score → Multi-domain profile
End-to-end information flow
Patient case History & exam
→
Reasoning Illness scripts, DDx
→
Metacognition Bias & gaps
→
Trust layer Calibrate & verify
→
Decision Spectrum options
→
Clinician Final judgment
↻ Learning — Longitudinal growth feeds back into next session
Paradigm shift
Conventional prompt engineering
GOAL Improve AI responses
MODEL AI generates conclusions
UNCERT Confidence statements only
DECIS Single recommended answer
LEARN Session-based, no continuity
Clinical Cognition Operating System
GOAL Improve human clinical reasoning
MODEL Clinician reasons; AI scaffolds
UNCERT Explicit epistemic calibration
DECIS Conservative → maximal spectrum
LEARN Longitudinal with structured reflection
Who this is for

Find your entry point

A Clinical Cognition Discovery Platform that executes reasoning frameworks, analytics modules, cognitive architectures, and agent pipelines on clinical cases to reveal the hidden processes of clinical thinking, decision-making, and metacognition.

Medical Students / Junior Doctors

Start at M01 – Socratic Reasoning. Use M04 before supervised ward rounds. Use M17 to sharpen your problem representation before any case discussion.

Family Caregivers / Advocates

Start at M02 – Patient Advocate Documentation, then M03 for ongoing monitoring. Use M11 when you need plain-language explanations of what is happening.

Clinicians in Practice

Use M05 to audit a live patient log. Use M09 for a complex case you want to write up. Use M13 when polypharmacy is a concern.

Educators / Prompt Authors

Use M08 to audit and quality-check any Socratic prompt you are writing or revising. Run Framework D at the end of any session to counter bias.

Important: All AI-generated outputs require independent clinical verification before being acted upon. Vibe Rounds is a patient-centred learning system — not a clinical decision support tool, prescribing aid, or diagnostic system. All outputs are learning observations, not clinical decisions.

Start with Module 0

New to Vibe Rounds? Let the cold-start orientation route you to the right module before any clinical content is entered.

Open Module 0 → Master Protocol Full Reference Repository ↗
Skills Library — Specialist Agents

57 reasoning modules

Each module is a self-contained prompt workflow with its own lifecycle (Initiation → Execution → Closure). Jump straight to the one that fits your session — or start at Module 0 if you're new.

⌕
57 modules
Meta-Skills — The Operating System

The pedagogical layer

Not standalone modules — woven into specific steps across the module set. These give every module access to the same underlying educational theory rather than each one inventing its own.

A

Humanistic Persona & Confidence-Building

Six traits that build clinical confidence alongside clinical competence. Key design rule: specific affirmation before challenge — naming the exact reasoning move, not generic praise. Challenge without affirmation triggers defensive cognition.

View framework →
B

Fink's Taxonomy of Significant Learning (FLINK)

Six non-hierarchical learning dimensions — foundational knowledge, application, integration, human dimension, caring, learning-how-to-learn — applied to keep every closure step producing durable, transferable insight, not just an answer.

View framework →
C

Bloom's Revised Taxonomy

Six cognitive levels (Remember → Understand → Apply → Analyse → Evaluate → Create), mapped explicitly to clinical reasoning tasks. Used to scaffold caregiver understanding and to calibrate the difficulty ratchet between sessions.

View framework →
D

Critical Awareness Framework

A standing closing prompt that names the cognitive biases the framework itself is susceptible to — automation bias, anchoring, hallucination risk, rare-diagnosis overweighting — with structured debrief prompts to counteract them. The protocol auditing itself, by design.

View framework →
Tutorial

How Guided Discovery Works

Guided Discovery is a Clinical Cognition Discovery Platform that reveals how clinical thinking develops around a case — not by generating answers, but by exploring how answers emerge.

The Core Idea

An MRI for Clinical Thinking

Most AI systems generate answers. Guided Discovery explores how answers emerge. Instead of asking a model to diagnose a case, you execute structured reasoning frameworks, cognitive architectures, analytics modules, and agent pipelines on clinical data.

Clinical cognition is usually invisible. You see diagnoses and decisions — but not the reasoning processes that produced them. Guided Discovery reveals structures, pathways, assumptions, biases, and reasoning strategies operating beneath the surface.

The Shift in Perspective
✕Most systems ask: "What is the diagnosis?"
✓Guided Discovery asks: "How does clinical thinking move from uncertainty to understanding?"
Three Levels of Exploration

Modules, Agents & Pipelines

Every run uses one of three building blocks — or combines them into a custom workflow.

Level 1

Modules

Single cognitive lenses providing focused insights into one aspect of a case.

Observation · Hypothesis Generation · Bias Detection · Decision Analysis
Level 2

Agents

Specialized cognition engines that orchestrate multiple frameworks in one run.

Guided Discovery Agent · Clinical Cognition Deep Dive · Analytics Agent
Level 3

Pipelines

Structured workflows combining modules and agents into progressive reasoning journeys.

Example: 1 → 12 → 9 → 21 → 35
The Discovery Journey

How a Case Unfolds

Running a pipeline takes a clinical case through progressive reasoning stages, each building on the last.

1
Observation
Raw case data
→
2
Pattern Recognition
Cue clustering
→
3
Hypothesis Generation
Differential building
→
4
Decision Architecture
Reasoning map
→
5
Bias Detection
Cognitive audit
→
6
Metacognitive Reflection
Cognition map
Ways to Run a Case

Worked Examples

Copy one of the queries below into Claude (or any AI) alongside a clinical case URL to start a run.

1

Analytics Mode

Run a single module. The AI acts as a cognitive assistant with reasoning rather than providing answers.

Query to Run
for case - https://classworkdecjan.blogspot.com/2019/05/42-f-with-severe-regular-edema-with_17.html?m=1 run module https://avi33tbtt.github.io/Prompts/Module-42-Clinical-Pre-Mortem.html
2

Targeted Clinical Pipeline

Chain multiple modules in sequence to walk a case through progressive reasoning stages.

Query to Run
for case - https://classworkdecjan.blogspot.com/2019/05/42-f-with-severe-regular-edema-with_17.html?m=1 run modules 1→ 12 → 9 → 21 → 35 from https://avi33tbtt.github.io/Prompts/
3

Full Guided Discovery Agent

Run the complete orchestrated agent — the deepest single-run exploration of a case.

Query to Run
for case - https://classworkdecjan.blogspot.com/2019/05/42-f-with-severe-regular-edema-with_17.html?m=1 run module https://avi33tbtt.github.io/Prompts/VibeRounds_Guided_Discovery_Agent.html
What Gets Revealed

What Can Be Discovered?

Each run surfaces a different layer of clinical cognition that is normally invisible in everyday practice.

Clinical reasoning pathways
Diagnostic strategies
Pattern recognition mechanisms
Decision architectures
Cognitive biases
Knowledge gaps
Uncertainty management
Metacognitive processes
Expert vs. novice thinking
Clinical expertise development
Prompt Creator

Build a run prompt

Select a case and chain modules in sequence — outputs a ready-to-paste prompt referencing the VibeRounds directory. Paste the result into any LLM.

⚡
VibeRounds Prompt Builder
Case → select modules → copy your run prompt
Case
Select modules
Run sequence
Select modules above to build your run sequence
Demo Implementations

See it run on real cases

Worked examples of the Master Protocol and individual modules applied to real case material — interactive dashboards, transcripts, and nested analyses.

ℹ️ How to view these: each demo lives on its own page. Click "Open demo" to launch it in a new tab — full interactivity, scrolling and printing all work best that way, rather than being squeezed into a frame on this page.
Flagship demos

The two most complete, end-to-end walkthroughs — start here if you only try two links.

Flagship · Analytics Dashboard

The Learning Stack — 14M Microcytic Anemia & Urinary Obstruction

A live, interactive run of the Master Protocol's outputs on a single pediatric case: case summary, top insights, prioritised investigations, differentials, and a printable action-plan report — all in one dashboard.

Case-based Interactive Printable report
Open demo ↗
Flagship · Case Reader

VibeRounds Case Reader

A guided, page-by-page reader for walking through a worked case transcript at a slower pace — useful for seeing how the Socratic questioning unfolds step by step rather than as a single dashboard.

Transcript view Step-by-step
Open demo ↗
More worked examples

Additional demos covering evaluation, tutorials, and nested cross-case analyses.

Analytics

The Learning Stack — Full Analytics Demo

The complete structured export covering every pedagogical framework, case reasoning step, and CARE report output.

22 Jun 2026Open ↗
Implementation

PaJR × Vibe Rounds — Prompts

Demo implementation pairing the Patient-as-Journal-Record (PaJR) approach with Vibe Rounds prompts.

19 Jun 2026Open ↗
Tutorial

CKD Class Tutorial

Socratic clinical reasoning walked through a chronic kidney disease class tutorial.

19 Jun 2026Open ↗
Case Practice

Clerkship Case Practice

Prompt templates for clerkship case practice, adapted for day-to-day ward use.

20 Jun 2026Open ↗
Nested Analysis

Nested Analysis — 5 EAI Cases

Lightweight nested analysis run across five EAI (early-AI-interaction) cases.

22 Jun 2026Open ↗
Nested Analysis

Nested Analysis — 4 Unrelated Cases

The same nested-analysis technique applied across four structurally unrelated cases.

22 Jun 2026Open ↗
Nested Analysis

Case vs. 5 PubMed Cases

A nested comparison of one original case against five matched PubMed case reports.

22 Jun 2026Open ↗
Research Papers & Preprints

Academic Publications

Research papers, technical architecture proposals, and preprints from the VibeRounds project — all available open-access on ResearchGate.

Preprint ★ Featured

VibeRounds: A Composable Architecture for Socratic Clinical Reasoning with Large Language Models — Design Rationale, System Description, and Composition Semantics for the Clinical Cognition Operating System

The flagship VibeRounds architecture paper — lays out the design rationale, system description, and composition semantics behind the Clinical Cognition Operating System underlying the framework.

↗ View on ResearchGate

Full Publication List

All preprints and works in progress are maintained on ResearchGate by Dr. Avinash Kumar Gupta.

↗ View ResearchGate Profile
CCOS at a Glance

Clinical Cognition Operating System — Summary

A one-page visual overview of how CCOS is positioned as a learning stack for clinical analytics and reasoning — not for decision making.

CCOS — Clinical Context Operating System: Learning Stack for Clinical Analytics and Reasoning Around a Case
↓ Download Image
Choose Your Goal. Follow the Pipeline. Master Clinical Thinking.
↓ Download Image
Vibe RoundsClinical Cognition OS (CCOS) — Module Order Builder, Learner Selection Sheet
FOR LEARNER USE
Architectural Note

From Prompt Library to CCOS Builder

The source directory began as a flat list of individual prompts, each written to exercise one clinical-reasoning skill in isolation. As the module set grew past 50 entries spanning reasoning, advocacy, analytics, safety, education, operations, and meta-design, the library outgrew "pick one prompt and run it." This builder is the next layer: it treats the modules as composable parts of a single prompt framework — a CCOS Builder — where the value comes from the order and combination of modules a learner chooses, not any one module alone.

Individual Prompts
→
Categorized Module Library
→
CCOS Builder: Ordered Pipelines

Concretely, that means three things this page adds on top of the raw module list: (1) tagging every module into one of seven cognitive lenses so a chain can be checked for breadth vs. depth, (2) preserving the exact sequence a learner selects modules in, since running "safety" before "reasoning" produces a different learning pass than the reverse, and (3) a plain-language read on what a given chain is likely to surface, so the choice of pipeline is itself a deliberate part of the exercise.

Not for Clinical Use

VibeRounds and this builder are a learning stack, not a clinical decision tool. Everything produced here — module chains, the analytic-yield commentary, and any pipeline you copy out — is a study aid for practicing clinical reasoning on deidentified, consented, publicly published cases, such as case reports. It is not validated for, and must not be used for, diagnosing or managing real patients. See the Disclosure Statement on the source site for full terms.

Tick modules in the order you'd run them on a case. Each tap stamps the next sequence number — untick and the chain renumbers itself. The panel on the right holds the pipeline exactly as chosen, plus a plain-language read on what kind of analytic yield that combination tends to produce.

Case Under Review

Provide the case the pipeline below will be run against — paste it in, add a link to where it lives, or both.
DUMMY CASE — REPLACE BEFORE USE

Chosen Pipeline

in selection order

    Readymade Pipelines

    Curated module chains for common situations — tap one to load it, then tweak in the builder above as needed.

    Beyond the module library

    Courses

    Three standalone courses that sit alongside the Vibe Rounds prompt stack — a full medical-curriculum directory, a from-scratch evidence-based-medicine course, and a from-scratch clinical-reasoning course. All three are free, self-paced, and built to be used with an LLM, not instead of thinking through the material yourself.

    🗂️

    Curriculum directory · Prompt Builder

    Courseware — the Ward Chart

    Open Courseware↗

    A complete index of the MBBS curriculum — 19 subjects, 190 chapters, pre-clinical through clinical — paired with a built-in Prompt Builder. You don't just browse chapters: you pick a subject, chapter, and topic (or switch to disease mode and pick straight from a disease list, or clinical-finding mode and pick a sign/symptom by body system), add one or more of 20 reusable prompt templates, and get a ready-to-copy prompt for any LLM.

    19Subjects
    190Chapters
    20Prompt templates
    3Browse modes

    Use it to:

    • Turn any chapter or topic into a Socratic case — commit to an answer first, earn hints, get the full answer only after you've reasoned it through.
    • Drill a single condition across every subject it touches, by switching to disease mode and picking it straight from the disease list — pathology, pharmacology, and clinical management in one pass.
    • Practice pattern recognition from the presentation backward, by switching to finding mode and picking a sign or symptom by body system — good for OSCE-style "here's a finding, build the differential" drilling.
    • Generate vignette MCQs, flashcards, exam Q&A, clinical pearls, or patient stories for any chapter, on demand, instead of hunting for pre-made question banks.
    • Rotate through the full curriculum in order — the directory is laid out as a rotation chart, pre-clinical → para-clinical → clinical — for structured, chapter-by-chapter self-study.
    📊

    9 lessons · Sequential

    Evidence-Based Medicine — from first principles

    Open EBM Course↗

    Critical appraisal taught the way a clinician actually thinks — starting with a patient, not a p-value. Nine lessons, meant to be done in order: each ends with homework that feeds the next, and the last lesson ties everything back together on one full patient case. There's also a Techie Summary for readers with no clinical background, and a built-in LLM prompt library for using an AI as a research assistant at each stage — with one hard rule throughout: the LLM drafts, you verify every number and citation.

    9Core lessons
    1Techie summary
    8Surprising-numbers insights

    Use it to:

    • Turn a vague clinical worry into a proper PICO question and a real, searchable PubMed strategy (Lesson 2).
    • Learn to appraise an RCT by hand — validity, results, applicability, plus ARR, RRR, NNT, and confidence intervals (Lesson 3) — and a systematic review or meta-analysis the same way, including reading a forest plot and checking AMSTAR-2 (Lesson 4).
    • Work through diagnostic-test studies (sensitivity, specificity, likelihood ratios, ROC curves, and how prevalence changes a test's usefulness) and prognosis/harm studies (cohort vs. case-control, hazard ratios, confounding) in Lessons 5–6.
    • Learn to read a clinical practice guideline critically (GRADE, industry influence) and reconcile it with a patient who doesn't match the trial average (Lesson 7).
    • If you have no clinical background, start with the Techie Summary — PICO as a search schema, RCTs as A/B tests, GRADE as test-coverage-vs-ship-decision — then use it as a running cheat sheet.
    • Use the LLM Prompt Library to draft a PICO question, build a search strategy, screen abstracts, or stress-test your own appraisal — always checked against the source.
    🧠

    13 lessons + 5 electives · Maps to the 57-module CCOS library

    Clinical Cognition — from first principles

    Open CC Course↗

    A dual-track course, for clinicians and technical readers alike, on how clinical reasoning actually moves — not just what conclusion it lands on. Lessons 1–9 build the core cognitive loop (observation → differential → bias-auditing → advocacy → analytics → safety → calibration → one full case run). Lessons 10–13 extend it into meta-cognition, systems and operations, precision medicine, and clinical wisdom. Every lesson names the exact CCOS modules (e.g. "M12, M15, M17…") it draws from, so you can jump straight from the lesson to the matching prompt.

    13Core lessons
    5Elective modules
    57Source CCOS modules

    Use it to:

    • Understand why the "AI that questions, not answers" philosophy works before using it — Lesson 2 lays out the full constraint set behind the Socratic Loop and why a forced first attempt changes what you retain.
    • Build a differential the disciplined way: four stages from raw findings to a stress-tested, probability-weighted differential, closing with a self-critique pass (Lesson 3).
    • Name your own blind spots — five angles on the same failure surface: the moment of commitment, the load-bearing finding, signal vs. noise, handoff distortion, and a full retrospective audit (Lesson 4).
    • Go beyond a single diagnosis with Lessons 10–13: meta-cognition, multi-agent systems and operations, precision-medicine personalization layers, and the closing habit of auditing your own compressed intuition.
    • Run the Master Protocol end to end on one case (Lesson 9) — the complete reasoning trace read back as a single artifact, checked for whether it actually holds together.
    • Cross-link into the EBM course wherever a lesson touches evidence — the two courses ask complementary questions ("how did I reason to this?" vs. "can I trust the evidence for it?") and can be started in either order.

    Using these courses with your learners

    None of the three courses require a live AI session to be useful on their own — but each becomes more powerful paired with a Vibe Rounds persona or a CCOS module run live on rounds, in a small group, or as a take-home assignment. A few concrete ways to slot them into teaching.

    🗂️

    With Courseware

    • Build a rotation-specific case set: pick the chapter your learners are on, generate vignette MCQs or Socratic cases via the Prompt Builder, and hand out a ready-made take-home problem set in minutes.
    • Run finding-mode stations for OSCE-style teaching — give the finding, not the diagnosis, and have learners build the differential live, the same way the tool is organized by body system.
    • Use disease-mode to assemble a cross-subject teaching session on one condition — pull the pathology, pharmacology, and clinical-management prompts for a single disease into one session.
    📊

    With the EBM Course

    • Assign one lesson as pre-reading, then spend group time on its homework — each lesson's homework is designed to feed directly into the next, so a weekly journal-club cadence maps onto the course almost as-is.
    • Use Lesson 9's full patient case as a capstone teaching session that ties PICO, search, appraisal, and decision back together in front of the group.
    • Hand non-medical or early-stage learners the Techie Summary first, so a mixed-background group can start the same session from a shared vocabulary.
    🧠

    With the CC Course

    • Use each lesson's explicit module mapping (e.g. Lesson 4 → M26, M28, M30, M37, M24) to pick exactly the right CCOS module for a specific teaching moment, instead of guessing which of 57 modules fits.
    • Start early learners on Lessons 1–3 (the Socratic Loop and differential-building) before ever putting them in front of a live persona prompt, so the "why" is in place before the "how."
    • Use Lessons 10–13 with senior residents or fellows — meta-cognition, systems/operations, precision medicine, and clinical wisdom sit well above the core loop most junior learners start with.

    A general pattern that works across all three: assign the relevant lesson or chapter as pre-work, then run the matching Vibe Rounds persona or CCOS module live — the course gives the "why" and the theory, the live run gives the practiced skill. Set the same ground rule either way: everyone commits to an answer before any hint or explanation is given.

    Questions about integrating these into a curriculum or study? Email avi33tbtt@gmail.com — happy to discuss.