WISE AI is an Agentic Clinical Decision Support System (CDSS) designed as part of the
TSAI – Extensive AI for Generative Systems Capstone Project.
It supports continued care across patients and doctors through consent‑based data capture, multi‑agent reasoning, and explainable outputs. It is designed to work strictly based on user-consent only, assisting both patients and doctors across the end‑to‑end healthcare journey — from triage and diagnostics to treatment guidance and continued care.
⚠️ This system provides decision support, not medical automation.
All clinical decisions remain with licensed practitioners.
The original capstone repository layout described in early TSAI materials (e.g. plugin/, cdss-app/, demo/, paper/) reflects the initial product vision and roadmap.
This README documents the current version of the repository: a FastAPI backend (backend/), a Vite + React web app (frontend/), Docker deployment under deployment/docker/, and the flows implemented today—including CBC and mental health workflows, Mock EHR APIs, optional S18 narrative support via long‑polling integration, and Supabase‑backed sign‑in for the WISE Clinical Portal.
The screenshots below show the current web experience: secure sign‑in, then the signed‑in WISE Clinical Intelligence Portal with Doctor/Patient workflow modes, operational profile (including S18 long‑polling and backend poll timeout messaging), and CBC workflow panels.
Sign in — WISE Clinical Portal (CBC and Mental Health workflows; Sign in / Create account / Forgot password)
Signed in — WISE Clinical Intelligence Portal (unified CBC and Mental Health workflows; optional S18 narrative support; Doctor vs Patient view)
To build a responsible, explainable, agentic healthcare assistant that supports — but never replaces — licensed clinicians.
WISE AI focuses on:
- Clinical decision support (not diagnosis)
- Patient‑centric continued care
- Human‑in‑the‑loop safety
- Transparent reasoning and confidence feedback
Design and demonstrate a production‑thinking AI system that:
- Uses agentic architecture (multiple cooperating AI agents)
- Is clinically responsible (decision support, not automation)
- Shows clear system design, reasoning, and UX
- Aligns with TSAI capstone dos & don’ts
Primary demo audience: Rohan Shravan (TSAI)
- A user‑invoked, consent‑based CDSS
- An assistive agent for optimal and efficient care delivery
- Agentic reasoning with confidence & feedback loops
- Designed for both patients and doctors
- Not an autonomous medical system
- Not silently monitoring users
- Not writing back to EHR automatically (future only)
These items reflect the conceptual architecture; the current codebase maps to backend/ (API, Mock EHR, orchestration), frontend/ (WISE Clinical Portal), and optional integration with S18 for extended narrative workflows.
-
WISE AI Plugin (vision / roadmap)
Browser‑based, user‑triggered, consent‑based extraction from EHR views; research and signal capture feeding a shared knowledge bank (not present as a separate top‑level package in the current tree). -
WISE AI CDSS Web App (current:
frontend/)
Standalone web UI for recommendations, clinical summaries, confidence scores, missing‑signal feedback, and orchestration of the agentic reasoning loop. Future or simulated actions are labelled accordingly. -
Shared Knowledge Bank
Research and RAG‑style context (platform‑level anonymised context, doctor and patient workspaces as concepts); supports feedback loops and iterative reasoning.
WISE AI follows a multi‑agent pattern, where each agent specializes in a specific reasoning task:
- Symptom Agent
- Lab / CBC Agent
- Trend & History Agent
- Research Agent
- Action / Recommendation Agent
All outputs are synthesized, scored for confidence, and presented for human approval.
📌 No agent directly writes to the EHR in MVP — all actions are advisory.
📐 Architecture diagrams
See under docs/architecture/ (for example docs/architecture/WISE_AI_CDSS_Architecture.md).
🚦MVP Feature Freeze (Capstone Scope) ✅ Included
- Manual invocation via “WISE AI” button in EHR
- Plugin‑based data capture (on demand)
- Multi‑agent reasoning
- CDSS UI with explainable outputs
- Confidence feedback loop
- Simulated future actions (clearly labelled)
❌ Explicitly Out of Scope (These are shown as future / disabled features where relevant)
- Automatic EHR write‑back
- Silent background monitoring
- Autonomous actions (lab booking, Rx ordering)
- Production compliance certifications
🧪 Demo Philosophy (for TSAI Evaluation)
- Real UI, real flows
- No mock screenshots passed as real
- Clear separation between:
- Working MVP logic
- Future extensibility
- Emphasis on agentic reasoning quality, not UI polish alone
🛠️ Technology Posture (current codebase)
- Frontend: React + Vite (
frontend/), Tailwind‑style UI for WISE Clinical Portal - Backend: FastAPI (
backend/), Mock EHR and CBC/orchestrator routes - AI layer: LLM‑driven agent orchestration (configurable)
- Local dev: Cursor IDE; optional Ollama for local models
- Demo LLM: Gemini (mentor‑preferred)
- Auth: Supabase (frontend); see frontend env and
deployment/dockerfor compose variables - Hosting (stretch): AWS (credits‑based)
- ⏱ 30‑day hard deadline
- 🎓 Academic capstone (design clarity > production scale)
- 🧪 PoC first, extensible architecture second
- 🧑⚕️ Clinical responsibility & explainability are non‑negotiable
- Sreedhar Byreeka — Product Manager & Healthcare IT SME
- Ritesh Verma — Agentic AI & Technical Lead
- Mentor: Rohan Shravan (TSAI)
🔄 Active development
📌 MVP feature set frozen
📐 Architecture finalised
🧪 Demo & evaluation in progress
Early capstone write‑ups described a layout similar to:
/
├── docs/
├── plugin/ # Browser plugin (data capture & research)
├── cdss-app/ # CDSS web app (UI + agents)
├── demo/
├── paper/
├── .github/
└── README.md
That structure reflects roadmap and narrative; the current repository layout below is what this project uses day‑to‑day.
/
├── backend/ # FastAPI app (Mock EHR, CBC, orchestrator)
├── frontend/ # Vite + React — WISE Clinical Portal
├── deployment/
│ └── docker/ # docker-compose, GHCR, full stack with S18 (see README there)
├── docs/ # Architecture, API notes, README images
│ └── images/ # Screenshots for this README
├── tests/ # Backend tests
├── data/
├── .github/
│ └── workflows/ # capstone.yml, backend-ci.yml, docker-publish.yml
└── README.md
Backend (from backend/):
pip install -r requirements.txt
uvicorn app.main:app --reload --host 0.0.0.0 --port 8000API docs: http://localhost:8000/docs
Frontend (from frontend/):
npm install
npm run devDev server uses port 5173 by default (vite.config.js). Configure API and Supabase env vars as needed for your environment (see frontend and root .env patterns used in Docker builds).
For building and running with Docker—local compose, images from GHCR, or full stack with S18Share—see deployment/docker/README.md.
The portal now enforces app roles:
doctor: can access CBC and mental-health clinical workflowspatient: cannot access doctor-only workflows
Set a provisioning secret in backend env:
DOCTOR_PROVISION_SECRET=change_me_long_random_valueGenerate a strong value:
python backend/scripts/generate_doctor_secret.pyPromote a user to doctor (replace values):
curl -X POST "http://localhost:8000/auth/provision-doctor" \
-H "Content-Type: application/json" \
-H "X-Provision-Secret: change_me_long_random_value" \
-d '{"user_id":"<supabase_user_id>"}'Check current role for a signed-in token:
curl "http://localhost:8000/auth/me" \
-H "Authorization: Bearer <access_token>"Workflows live under .github/workflows/, including:
capstone.yml— basic repository checksbackend-ci.yml— backend CIdocker-publish.yml— container image publishing
WISE AI is a clinical decision‑support system.
It does not diagnose, prescribe, or replace licensed medical professionals.
All outputs are advisory and require human clinical judgment.
📚 References
- TSAI – The School of AI: https://theschoolof.ai/
- Course: Extensive AI for Generative Systems
- Capstone Guidelines: See
/docsand uploaded PDFs