Kanishk Paul

Projects

Agents, reasoning tools, local models, and sharp experiments.

Public projects from my GitHub and resume, focused on browser automation, interpretability, computer-use agents, offline inference, and AI-assisted chess analysis.

TypeScript / Playwright

ChromeClaw

Terminal-first browser agent with structured observation, JSON planning, Playwright execution, traces, evals, and safety gates.

React 19 / TypeScript

ReasonTrace

Reasoning-trace debugger with transcript ingestion, graph rendering, timeline scrubbing, diagnostics, export, and replay.

Python / Flask

Casca

Screenshot-based desktop agent scaffold around an observe-propose-validate-execute loop with bounded, replayable runs.

Rust / llama.cpp

Sefai

Offline GGUF inference CLI that runs local models through llama.cpp bindings with CPU-first defaults and optional accelerators.

TypeScript

Stockfih

Chess analysis with Stockfish plus language-model coaching, connecting structured engine output to more useful learning loops.