jaeyoonwang/rwanda-cdss

Rwanda CDSS Pilot (Malaria + HIV) # Malaria & HIV Clinical Decision Support System A multi-agent CDSS that answers clinical queries using Rwanda national guidelines for **Malaria** (2024) and **HIV**. An LLM-powered orchestrator routes queries to the appropriate specialist agent(s) and synthesizes responses — including drug interaction warnings for coinfection scenarios. > For full engineering details, architecture diagrams, API reference, and deployment guide, see ENGINEERING_HANDOFF.md. --- ## Architecture ``` User / REST API ↓ CDSSOrchestrator ├── QueryRouter (GPT-4o) → MALARIA | HIV | BOTH | UNCLEAR ├── clinical_map.json → coinfection triggers, drug interaction flags │ ↓ ↓ │ MalariaAgent HIVAgent │ (malaria KB) (hiv KB) └── ResponseSynthesizer → merged answer + interaction warnings ``` Each guideline agent uses a 6-stage PDF processing pipeline: ``` Raw PDF ├─ Stage 1: PDF Parser (pdfplumber / Claude Vision for image-based PDFs) ├─ Stage 2: Section Builder (rule-based heading detection) ├─ Stage 3: Figure Extractor (Claude Vision — flowcharts, decision trees) ├─ Stage 4: Table Structurer (Claude Vision — dosage tables, regimen matrices) ├─ Stage 5: Prose Chunker (Claude Sonnet — atomic clinical fact chunks) └─ Stage 6: KB Assembler (drug normalization, eligibility matrix) ``` --- ## Quick Start ### 1. Prerequisites ```bash # Python 3.10+ pip install -r requirements.txt # macOS (for HIV pipeline image rendering) brew install poppler # Linux apt-get install poppler-utils ``` ### 2. Environment ```bash cp .env.example .env # Edit .env and fill in your API keys: # OPENAI_API_KEY=sk-... # ANTHROPIC_API_KEY=sk-ant-... ``` ### 3. Set Up the Malaria Knowledge Base The malaria KB was generated by the existing pipeline. Copy it to the new location: ```bash mkdir -p guidelines/malaria/output cp output/*.json guidelines/malaria/output/ ``` ### 4. Run the HIV Pipeline (one-time setup) > **Cost wa …