AI-generated trading strategies and Risk Management for hedge funds and asset managers—automatically.
Built by AGVEST Technologies Private Limited
Traditional quant funds still rely on human researchers hand-crafting signals. Niveshi replaces that cycle with a reinforcement-learning (RL) engine that invents, validates and deploys new intraday strategies with zero manual bias—letting us iterate roughly 100 × faster than legacy shops.
Agents decide actions directly (long, short, flat) rather than predicting prices, enabling full automation.
One engine ports to any asset class; multiple equity sectors already live.
Strategies get better as more data arrives, instead of decaying like hand-tuned models.
Every trade carries a confidence score; our allocator funnels capital only to the highest-edge signals.
Minute-level OHLCBA data streams in; RL agents learn by reward while an input-filtering layer removes irrelevant stocks automatically.
Strategies must pass strict quant & ML filters (draw-down limits, noise sensitivity, regime robustness) before deployment.
Live signals feed a confidence-weighted allocator that selects the most profitable trades every tick.
| Step | What Happens | Tech Highlight |
|---|---|---|
| Data ingestion | Continuous minute-bars streamed, denoised, normalised | Auto feature-engineering inside RL network |
| RL engine | Cloud swarm of agents competes to maximise reward | Novel bucket-sampling prevents regime over-fit |
| Execution loop | Allocator deploys high-confidence orders; fills pipe back as new training data | Feedback loop keeps models current |
Ex-Accredible engineer; open-sourced deep_trader Link RL project; oversees infra & cloud engine.
Former Quantmetrics London quant; automated their execution desk; leads research pipeline.