This project is an agentic healthcare automation system that orchestrates end-to-end clinical workflows using multimodal AI, vector-based medical retrieval, and secure EMR actions. It reduces physician administrative load by automating listening, summarization, symptom extraction, investigation planning, prescription drafting, and pharmacy order dispatch.
The system integrates:
- Live camera feed + facial recognition checkpoint
- Locks all clinical actions until patient identity is confirmed
- Prevents unauthorized EMR or prescription access
- Acts as a “Presence Agent” in the workflow
- Supports live microphone streaming + WAV audio upload
- Uses robust system-level speech-to-text (non-cloud)
- Converts consultation speech → structured transcript
- No dependence on paid APIs
A chain of domain-specific agents built as modular nodes:
-
Scribe Agent → cleans transcript, produces clinical note
-
Symptom Extraction Agent → maps lay language → canonical symptoms
-
Planner Agent → recommends investigations via
- rule-based mapping
- local Qdrant vector DB + guideline retrieval
-
Prescription Agent → drafts structured provisional medication plan
-
Safety Agent → identifies emergency red flags
These nodes form an Agentic Workflow Graph orchestrated by the backend.
- Local Qdrant Vector DB stores medical guideline chunks
- Planner agent performs guideline-aware test selection
- No external LLMS used
- Fully offline & reproducible
- After doctor review, approved consultations are persisted into:
emr_store.json→ mock EMR system - All EMR actions enforce biometric gate
- Every state transition is logged in the audit trail
- Converts approved prescriptions → pharmacy orders
- Writes orders to
pharmacy_orders.json - Includes EMR record linkage, timestamping, and action metadata
- Demonstrates real-world “agent → external tool” interoperability
The backend serves an interactive, agent-aware dashboard with:
- Face verification panel
- Audio/Livemic clinical capture
- Editable transcript
- Auto-generated symptoms, tests, prescription
- EMR records viewer
- Pharmacy orders viewer
- Audit timeline for explainability
Everything is rendered using pure HTML/CSS/JS inside FastAPI—no external frontend frameworks needed.
- Agentic workflow automation for clinical operations
- Secure patient-gated EMR write access
- Healthcare RAG engine using local vector database
- Multimodal STT ingestion pipeline
- Explainable AI with step-by-step audit logs
- End-to-end consultation → EMR → Pharmacy flow
- Completely offline, no paid cloud APIs
- Modular nodes enabling extensibility (labs, insurance, billing, alerts)