Skip to content

DEPLOYMENT.md

Latest commit

 

History

History
73 lines (54 loc) · 2.24 KB

DEPLOYMENT.md

File metadata and controls

73 lines (54 loc) · 2.24 KB

Deployment Guide

Split deployment: FastAPI backend on Render, Vite React frontend on Vercel.

Backend — Render

  1. Create a new Web Service on render.com, pointing at this repo.
  2. Render auto-detects render.yaml — no manual settings needed.
  3. Add the required environment variable in the Render dashboard:
    • OPENAI_API_KEY — your OpenAI API key
  4. The service exposes a /health endpoint used by Render's health check.
  5. WebSocket connections are served at /ws on the same service.

Start command (from render.yaml):

uvicorn src.api.main:app --host 0.0.0.0 --port $PORT

Note: stable-baselines3, torch, and torch-geometric increase build time significantly. Use the Standard plan or above for sufficient build memory.

Frontend — Vercel

  1. Import this repo on vercel.com.
  2. In project settings, set Root Directory to src/ui/dashboard.
  3. Framework preset: Vite (auto-detected).
  4. Add environment variables:
    • VITE_API_URL — your Render service URL, e.g. https://cybercypher-api.onrender.com
    • VITE_WS_URL — (optional) WebSocket URL. If omitted, derived automatically from VITE_API_URL (https → wss, appends /ws).
  5. vercel.json in src/ui/dashboard/ handles SPA rewrites — no extra config needed.

Local Development

Backend:

cp .env.example .env
# fill in OPENAI_API_KEY
pip install -r requirements.txt
uvicorn src.api.main:app --reload

Frontend:

cd src/ui/dashboard
cp .env.example .env.local
# set VITE_API_URL=http://localhost:8000 (or leave blank; vite.config.js proxies /api and /ws)
npm install
npm run dev

RL Training (optional, improves DeciderAgent)

python train_rl_synthetic.py --timesteps 50000 --output models/rl_traffic_engineer

The model is auto-loaded by DeciderAgent at startup from models/rl_traffic_engineer.zip.

LLM Fine-tuning (optional, domain-specific incident reasoning)

# Step 1 — generate synthetic incident dataset
python src/models/llm_finetune/synthetic_incident_generator.py --count 500 --output data/incidents.jsonl

# Step 2 — fine-tune with LoRA
python src/models/llm_finetune/train_lora.py --dataset data/incidents.jsonl