humanovo — Adversarial Biomedical Hypothesis Engine

An AI-native platform for biomedical hypothesis generation using a 12-stage adversarial multi-model pipeline with dual-embedding grounding. Every hypothesis is generated, attacked, revised, mechanistically validated, and scored before delivery — with full citation provenance and audit trails.

Architecture

humanovo/
├── backend/                  # Python FastAPI backend
│   ├── app/
│   │   ├── agents/           # 12-stage discovery orchestrator + supporting agents
│   │   ├── api/v1/           # REST API + WebSocket endpoints
│   │   ├── compute/          # Domain compute (genomics, pharma, imaging, signals)
│   │   ├── entity_resolution/# Canonical ID resolution + synonym management
│   │   ├── etl/              # Bulk data loading + dataset parsers
│   │   ├── ingestion/        # Data ingestion pipeline (PubMed, trials, omics)
│   │   ├── integration/      # Neo4j graph + pgvector + provenance connectors
│   │   ├── knowledge/        # Knowledge engine (graph + vector hybrid)
│   │   ├── literature/       # Literature pipeline (criteria, snapshots, updates)
│   │   ├── models/           # SQLAlchemy data models
│   │   ├── nlp/              # NLP pipeline (NER, relation extraction, assertion)
│   │   ├── rag/              # RAG pipeline (chunker, embeddings, retriever, reranker)
│   │   ├── scoring/          # Citation analysis + claim classification + confidence
│   │   ├── services/         # Business logic services
│   │   └── simulation/       # Monte Carlo simulation engine
│   └── tests/
├── frontend/                 # React TypeScript frontend (Vite + Tailwind)
├── infrastructure/           # Terraform (AWS)
├── docker/                   # Docker configurations
└── scripts/                  # Deployment + data loading utilities

12-Stage Discovery Pipeline

Each hypothesis passes through 12 specialized LLM stages sequentially. Between EVERY stage, dual-embedding grounding verifies claims against evidence.

Stage	Role	Model	Provider
1. SEED	Generate initial hypothesis	Claude Opus 4.6	AWS Bedrock
2. EXPAND	Broaden hypothesis scope	Claude Sonnet 4.6	AWS Bedrock
3. EVIDENCE	Literature evidence review	Cohere Command A	Azure OpenAI
4. COUNTER	Adversarial counter-arguments	Mistral-Large-3	Azure AI
5. REVISE	Revise based on counter-arguments	o3-mini	Azure OpenAI
6. MECHANISM	Mechanistic deep dive	GPT-4.1	Azure OpenAI
7. VALIDATE	Cross-validation	Claude Sonnet 4.6	AWS Bedrock
8. GROUND	3-layer scientific grounding	Grok-4-1-fast	Azure AI
9. SCORE	Multi-dimensional confidence	GPT-4.1	Azure OpenAI
10. REFINE	Fast refinement	GPT-4o	Azure OpenAI
11. TRANSLATE	Translational roadmap T0-T5	Claude Sonnet 4.6	AWS Bedrock
12. FINALIZE	Final synthesis	Claude Sonnet 4.6	AWS Bedrock

Dual-Embedding Grounding (between every stage)

Two embedding models run in parallel on every stage output:

1. Bedrock Cohere Embed English v3 (1024d) — biomedical-optimized
2. Azure text-embedding-3-large (1536d) — general-purpose

Grounding mechanisms:

• RAG Retrieval: Embed output → retrieve matching evidence → inject into next stage
• Semantic Gating: Compare each claim against evidence pool → flag ungrounded claims

Data Sources (60+ APIs)

Core: PubMed, ClinicalTrials.gov, openFDA, UniProt, Reactome, KEGG, Ensembl, HMDB Extended: Elsevier/Scopus, Springer Nature, ChEBI, HCA, NCBI Gene, ClinVar, Semantic Scholar, OpenAlex, ChEMBL, DrugBank, DisGeNET, STRING, PDB, AlphaFold, WikiPathways, and more.

Tech Stack

Backend: Python 3.11+ · FastAPI · SQLAlchemy + asyncpg · Neo4j · pgvector · Celery + Redis Frontend: React 18 + TypeScript · Vite · TanStack Query · Zustand · Cytoscape.js Infrastructure: AWS (Terraform) · Docker · PostgreSQL · Redis LLM Providers: AWS Bedrock · Azure OpenAI · Azure AI Foundry

Quick Start

Prerequisites

• Docker and Docker Compose
• Node.js 22+
• Python 3.11+
• API keys for at least one LLM provider (AWS Bedrock or Azure OpenAI)

Setup

# Clone
git clone <repository-url>
cd humanovo

# Copy and configure environment
cp backend/.env.example backend/.env
# Edit .env with your API keys

# Start infrastructure
docker-compose up -d postgres redis neo4j

# Backend
cd backend
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt
alembic upgrade head
uvicorn app.main:app --reload

# Frontend (new terminal)
cd frontend
npm install
npm run dev

Run Pipeline Integration Test

cd backend
pytest tests/integration/test_pipeline_e2e.py -v --timeout=600

License

Proprietary — All rights reserved. © 2025-2026 Adyanthaya Ventures.