AKASH Kumar

At IntelGrader, I shipped full-stack features to client-facing production environments and contributed to architectural decisions on the production codebase. I also built a text-to-video pipeline that turns one prompt into a narrated 2D educational video using validated JSON scene plans and automated rendering.

On Chronos, my reactive market simulator, I wrote a single-threaded matching engine with strict price-time priority and SQLite persistence so seeded runs replay exactly. I separated order entry from market data and added a FIX 4.4 gateway and sandboxed strategy SDK.

I engineered time-series features from one-minute OHLCV data and trained PPO agents under transaction costs, slippage, volume caps, and drawdown penalties. A pytest suite checks book priority, ledger conservation, book-driven prices, and deterministic runs.

On Reflow, I benchmarked a payment recovery agent against four baselines using seeded randomness and an injected clock; in a simulated 200-event batch, it recovered INR 5.73 lakh with 112 charge attempts. I also built PatternIQ, a strategy-building and backtesting platform with a sandbox for generated code, plus news NLP, anomaly detection, and candlestick-pattern models.