Mednexa: Pharma Portfolio Planning Platform (Agentic AI)

Mednexa is a full-stack platform for pharmaceutical portfolio planning, combining a Python-based agentic backend (LangGraph, Gemini, FastAPI) with a modern Next.js/TypeScript frontend. It orchestrates specialized worker agents, aggregates insights, and generates executive PDF reports.

---

System Architecture

Folder Structure (Backend & Frontend)

MedNexa2/
├── app.py                # Backend entry point (CLI)
├── api.py                # FastAPI backend server
├── pyproject.toml        # Python dependencies
├── agents/               # All agent logic (IQVIA, EXIM, Patent, etc)
├── orchestration/        # LangGraph state machine, router, state
├── contracts/            # Pydantic schemas for agent IO
├── data/                 # Mock data files (JSON, CSV)
├── reports/              # PDF generation and templates
├── llm/                  # Gemini summarizer
├── outputs/              # Generated PDF reports
├── mednexa-frontend/     # Next.js 14 frontend (TypeScript, Tailwind)
│   ├── src/app/          # App Router pages (dashboard, chat, results, etc)
│   ├── components/       # UI, layout, chat, charts
│   ├── lib/              # Utilities
│   └── services/         # API service functions

LangGraph Node Graph (Backend Flow)

START
	↓
[Parse User Query]
	↓
[Master Agent: Query Analysis]
	↓
[Rule-Based Router] ← Determines which agents to invoke based on query keywords
	↓
	├─→ [IQVIA Agent] ──┐
	├─→ [EXIM Agent] ────┤
	├─→ [Patent Agent] ──┤
	├─→ [Clinical Trials Agent] ──┤
	├─→ [Internal Knowledge Agent] ──┤
	└─→ [Web Intelligence Agent] ──┘
					↓
	[Aggregator Node] ← Collects all worker outputs into single JSON
					↓
	[Gemini Summarizer Node] ← **ONLY LLM CALL** (receives aggregated JSON)
					↓
	[PDF Report Generator]
					↓
	[END: Print file path]

---

Backend Usage

Command Line

python app.py "What is the market potential for Drug X in oncology?"

Interactive Mode

python app.py
# Then enter your query when prompted

API Server

uvicorn api:app --reload

POST to /analyze with { "query": "..." } for summary and PDF.

Environment Variables

• GEMINI_API_KEY: Required for Gemini summarization

Output

• PDF reports are saved to outputs/ directory
• Console displays processing status and executive summary

Data

Mock data files are in data/:

• iqvia_data.json, exim_data.csv, patent_data.json, clinical_trials_data.json, internal_knowledge.json, web_intelligence.json

---

Frontend Usage (mednexa-frontend)

Prerequisites

• Node.js 18+, npm or yarn

Installation

cd mednexa-frontend
npm install

Development

npm run dev
# App at http://localhost:5000

Production

npm run build
npm start

Backend Integration

Set backend URL via NEXT_PUBLIC_API_URL or in src/services/api.ts.

---

Demo Flow (End-to-End)

1. Dashboard → View KPIs and recent queries
2. Chat → Ask "Where is the unmet need in oncology?"
3. Clarify → Select region or mechanism focus
4. Results → Explore detailed agent outputs
5. Summary → View executive report with findings
6. Download → Generate PDF report

---

Constraints & Design Principles

• Worker agents do NOT use LLMs
• Gemini is called exactly ONCE at the end
• All routing is rule-based (keyword matching)
• All inter-agent communication uses structured JSON
• Frontend and backend are fully decoupled (API-driven)

---

License

Private - Mednexa