No Average Trading is a two-part system. The Next.js app is the control plane — dashboard, trade execution, cron jobs, Alpaca + Supabase integration. The Python subproject underpython/ is a full reinforcement-learning stack (gym environment, PPO trainer, Optuna tuner, FastAPI inference server) that trains a policy on Modal and serves weight vectors back to Next.js through a shared Bearer-token endpoint.

Core Workflow

Every weekday, the data-sync cron pulls OHLCV into Supabase, then the rebalance cron asks the selected strategy for target weights, computes deltas vs current positions, and sends GTC market orders to Alpaca. Only deltas larger than 2% trigger trades, keeping turnover costs in check.