RL-powered trading research for Indian equities (NSE & BSE).
RLAlphaLabs is a solo research initiative applying deep reinforcement learning to quantitative trading on Indian markets — built with a focus on realistic costs, rigorous validation, and honesty about what actually works, rather than overfit backtests or black-box hype.
Live site: https://pctablet505.github.io/RLAlphaLabs/
- Coverage: 700+ NSE stocks, 6 market indices, across 8 time granularities
- Data: 5,000+ processed data files powering training and evaluation
- Agents: actor-critic RL over a discrete, multi-slot portfolio action space, implemented in JAX and trained across many vectorized environments
- Validation: walk-forward testing on held-out data (not just backtests), plus adversarial stress tests and a paper-trading gate — not single-run backtest metrics
- Costs modeled explicitly: NSE/BSE fees, STT, and other transaction costs are built into the simulation, not ignored
- Reward design: risk-aware rewards rather than raw PnL maximization
- GPU-accelerated training
This project is in an active research phase — not a live trading product. There is no live capital deployed and no performance guarantee. See the Research page for published results and methodology, and About for the project's goals and principles.
This repository hosts the public GitHub Pages site for RLAlphaLabs (features, research write-ups, and experiment logs). The underlying trading codebase — data pipelines, environments, and training infrastructure — is private; this site is the public window into that work.