DOCKER.md

AILIXIR Docker Stack

Monorepo stack: Laravel API + ADMET + Drug Repurposing + Chemical RAG.

Requirements

• Docker 24+ with Compose v2
• ~16 GB RAM recommended (PyTorch + ADMET models)
• Model artifacts present locally:
  • ai_apps/ADMIT/admet_inference/models/*/best_model.ckpt
  • ai_apps/Drug Reporposing/save_folder/pretrained_models/...

Quick start

# From repository root
docker compose build
docker compose up -d

Service	Host URL
Laravel API	http://localhost:8080
ADMET	http://localhost:8002
Drug Repurposing	http://localhost:8001
Chemical RAG	http://localhost:5000

Laravel → AI integration endpoints

Enabled when AI_INTEGRATION_ROUTES_ENABLED=true (set in docker/laravel.env).

Method	Endpoint	Description
GET	/api/ai-services/health	Ping all AI /health endpoints
POST	/api/ai-services/test/admet	Proxy ADMET /predict
POST	/api/ai-services/test/chemical-search	Proxy Chemical RAG retrieval
GET	/api/ai-services/test/drug-repurposing	Health + model status

Example:

curl http://localhost:8080/api/ai-services/health
curl -X POST http://localhost:8080/api/ai-services/test/admet \
  -H "Content-Type: application/json" \
  -d '{"smiles":"c1ccccc1"}'

Images

• Laravel: multi-stage (composer + node + php:8.3-cli-alpine), runs as www-data
• ADMET / Drug Repurposing: existing multi-stage Python slim images, non-root UID 1000
• Chemical RAG: multi-stage python:3.11-slim, non-root raguser
• MySQL: mariadb:11.4

CI

GitHub Actions workflow: .github/workflows/integration-test.yml

Builds the full stack and tests Laravel proxy endpoints against all AI services.

Environment

Copy and adjust docker/laravel.env for container-specific values. For local (non-Docker) development, set URLs in .env using .env.example as reference.