- Building TCalc — a local-first token intelligence toolkit for coding agents and model selection.
- Building SpeciesBench — a benchmarking framework to measure speciesist bias in LLM responses.
- Exploring multilingual LLM safety with IndiaJailbreakBench-Lite.
- Shipping real-world full-stack products like FindXVision.
| Project | What it does | Core Stack |
|---|---|---|
| TCalc | Scans codebases for token usage, recommends suitable models, and generates agent-ready context maps | TypeScript · VS Code Extension · CLI · MCP |
| SpeciesBench | Measures speciesist bias signals in LLM outputs with reproducible scoring | Python · Streamlit · SQLite · OpenRouter |
| IndiaJailbreakBench-Lite | Evaluates jailbreak/refusal consistency across Indian languages | Python · LLM Evaluation |
| FindXVision | Missing-person platform with case tracking and face-recognition workflow | MERN · Redux · Twilio |
| Academic Research Tracker | Prompt-driven academic alerting system with automation workflows | Next.js · PostgreSQL · n8n |
Languages: Python, TypeScript, JavaScript, Java, C/C++
Frontend: React, Tailwind CSS, Vite
Backend: Node.js, Express.js, REST APIs
Data & AI: SQLite, PostgreSQL, Streamlit, OpenRouter, Groq
Tools: Git, Docker, GitHub Actions, Linux
