AgentRX

Overview

AgentRX is a MedTech / Agentic-AI platform for accelerating pharmaceutical drug repurposing discovery. It orchestrates autonomous worker agents, manages stateful multi-step pipelines, and produces structured decision outputs rather than chat-based responses.

The current repository contains a backend-first implementation with:

• FastAPI for API endpoints
• LangGraph for stateful orchestration
• Celery / Redis for async task handling
• Local mock enterprise data for IQVIA and EXIM analysis
• A resumable pipeline with human approval pause points

---

Architecture

graph TD
    U[User / Frontend] -->|POST /api/pipeline/start| API[FastAPI Backend]
    API -->|start pipeline| LG[LangGraph Orchestrator]
    LG --> WI[Web Intelligence Agent]
    LG --> PL[Patent Landscape Agent]
    LG --> CV[Commercial Viability Agent]
    LG --> IQ[IQVIA Supply Chain Agent]
    WI -->|PubMed| PUB[PubMed Entrez API]
    PL -->|Europe PMC| EPC[Europe PMC REST API]
    CV -->|OpenFDA| FDA[OpenFDA API]
    CV -->|ClinicalTrials.gov| CT[ClinicalTrials.gov API]
    CV -->|Yahoo Finance| YF[Yahoo Finance]
    IQ -->|Local CSV| CSV[Mock IQVIA/EXIM CSV Data]
    API -->|checkpoint| SQLITE[SQLite checkpoint file]
    API -->|stores reports| MINIO[MinIO / S3]
    API -->|optional| REDIS[Redis broker]
    API -->|optional| POSTGRES[PostgreSQL]
    API -->|optional| CHROMA[ChromaDB]

Loading

---

Repo Structure

AgentRX/
├── backend/
│   ├── app/
│   │   ├── agents/          # Worker agents and business logic
│   │   ├── api/             # FastAPI route definitions
│   │   ├── core/            # Configuration, Celery setup
│   │   ├── pipeline/        # LangGraph workflow and state definitions
│   │   ├── models/          # Pydantic models and DB models (planned)
│   │   └── services/        # External API clients and helpers
│   └── templates/           # Jinja2 report templates
├── config/                  # Pipeline/config metadata
├── data/                    # Persistent data, mock CSVs
├── frontend/                # Frontend integration placeholders
├── infrastructure/          # Docker and deployment assets
├── mock__db_services/       # Local mock API services
│   ├── exim/
│   └── iqvia/
├── requirements.txt
├── docker-compose.yml
├── .env.example
├── context.md
└── README.md

---

Current System Capabilities

1. Pipeline Orchestration

The backend exposes a resumable decision pipeline via:

• POST /api/pipeline/start
• POST /api/pipeline/resume

The pipeline currently executes:

1. Pharmacodynamic Mapping via the Web Intelligence Agent
2. Data Merge into merged_diseases
3. IP Whitespace Clearance via the Patent Landscape Agent
4. Commercial Viability Analysis via the Commercial Viability Agent
5. IQVIA / EXIM Supply Chain Analysis via the IQVIA Supply Chain Agent

A manual approval pause occurs before the commercial viability phase.

2. Agent Implementations

Agent	Role	Data Sources	Current Status
Web Intelligence Agent	Extracts alternative indications from literature	PubMed Entrez + Gemini LLM	Implemented
Patent Landscape Agent	Filters by active patents / FTO	Europe PMC	Implemented
Commercial Viability Agent	Estimates TAM, competition, trial complexity	OpenFDA, ClinicalTrials.gov, Yahoo Finance, OpenRouter LLM	Implemented
IQVIA Supply Chain Agent	Reads local IQVIA / EXIM CSV data, builds supply chain output	Local mock CSVs	Implemented

---

How to Run Locally

Prerequisites

• Python 3.11+ (recommended)
• Docker / Docker Compose
• Git

Setup

1. Create a virtual environment:

python -m venv .venv
source .venv/Scripts/activate    # Windows
source .venv/bin/activate        # macOS/Linux

2. Install dependencies:

pip install -r requirements.txt

3. Copy environment variables:

cp .env.example .env

4. Update .env with your API keys and service endpoints.

Run services with Docker Compose

docker-compose up -d

The compose stack includes:

• postgres for persistent relational storage
• redis for Celery broker/backend
• chromadb for vector storage
• minio for report storage

Start the Backend

cd backend
uvicorn app.main:app --reload

Run the Pipeline

Use the /api/pipeline/start endpoint to begin and /api/pipeline/resume to continue paused runs.

---

API Endpoints

POST /api/pipeline/start

Starts the pipeline for a molecule.

Request

{
  "molecule": "Metformin"
}

Success Response

{
  "status": "PAUSED_FOR_HUMAN",
  "thread_id": "<uuid>",
  "message": "Awaiting human approval for IP-cleared candidates.",
  "pending_candidates": []
}

POST /api/pipeline/resume

Resumes a paused pipeline run.

Request

{
  "thread_id": "<uuid>"
}

Success Response

{
  "status": "COMPLETED",
  "molecule": "Metformin",
  "final_candidates": [],
  "iqvia_exim_analysis": {}
}

GET /health

Health check endpoint for the backend.

Response

{
  "status": "online",
  "environment": "development",
  "orchestrator": "LangGraph Ready",
  "message": "AgentRX Neural Backend is operational"
}

---

Data & Mock Services

Mock Services

The repository includes local mock services for enterprise data:

• mock__db_services/iqvia/ — mock IQVIA market-insights API
• mock__db_services/exim/ — placeholder for EXIM trade data services

Local CSV Data

Key mock data files are stored in data/mock_seeds/:

• IQVIA_DrugProfiles.csv
• IQVIA_Sales.csv
• IQVIA_ClinicalPipeline.csv
• EXIM_CountryMatrix.csv
• EXIM_TradeData.csv

These files are consumed by the IQVIA Supply Chain Agent to produce supply chain summaries.

---

Configuration

The backend uses app/core/config.py to load environment variables from .env. Important config keys include:

• POSTGRES_URL
• REDIS_URL
• CHROMADB_HOST
• CHROMADB_PORT
• NEO4J_URI
• GOOGLE_API_KEY
• OPENROUTER_API_KEY
• S3_ENDPOINT_URL
• AWS_ACCESS_KEY_ID
• AWS_SECRET_ACCESS_KEY

---

Known Limitations

• PostgreSQL and Neo4j integrations are referenced in config, but the current pipeline uses SQLite checkpoints for runtime state.
• RAG vector search and Neo4j knowledge graph features are planned, not yet wired into the current flow.
• Some external API calls are mocked or dependent on available API keys.

---

Development Notes

• The pipeline is defined in backend/app/pipeline/graph.py using LangGraph state nodes.
• Worker agent logic lives in backend/app/agents/workers.py.
• The pipeline REST API lives in backend/app/api/routes/pipeline.py.
• backend/app/core/celery_app.py configures Celery, but task modules are minimal.

---

Contribution

1. Fork the repo
2. Create a feature branch
3. Run the backend locally and test changes
4. Submit a PR with clear architectural notes

---

Contact

For this repository, use the internal context.md as the source of truth for architecture, current implementation state, and feature planning.