AI Tech Lead · LLM Infrastructure · Agent and Model Evaluation · AI for Public Education
I build LLM infrastructure and evaluation tools, and I lead the AI engineering behind state education systems in India that reach 66,000 government teachers.
Portfolio · LinkedIn · [email protected]
| Project | What it does | Built with |
|---|---|---|
| MultiLLM Proxy | Self-hostable LLM gateway: one OpenAI- and Anthropic-compatible API in front of 25 providers, with scoped keys, safe failover between providers, streaming, rate limits, and cost and latency telemetry. Source-available | Python, Flask, Cloudflare Workers |
| CounterLab | Belief debugger for ML evaluation mistakes. You state a claim and lock a prediction, a controlled experiment runs on your notebook's evidence, and a frozen verifier decides whether the result counts. Covers data leakage and class imbalance | TypeScript, Jupyter, Cloudflare Workers |
| Omni (private) | Personal agent harness: a multi-provider model router, tool use under approval policies, long-term memory, scheduled jobs, audit, and an eval harness | TypeScript, Cloudflare Workers, D1 |
| VeriLearn | Turns YouTube lecture playlists into evidence-backed exam prep: timestamped notes, question banks, mock exams, and mastery tracking | Next.js, Gemini API |
| LayerForge-X | Depth-aware, amodal decomposition of a single RGB image into an editable layer graph | Python, computer vision |
| image-and-video-studio | Local-first studio for AI image, video and text-to-speech generation across providers | TypeScript, Next.js |
Open source: I maintain the Event Calendar widget for KDE Plasma 6 (★100), a calendar, agenda and weather plasmoid with Google Calendar sync and iCalendar import.
As AI tech lead, I lead the team that designs, builds and rolls out AI and data systems with state education departments. The work began in Himachal Pradesh and is expanding to multiple states. These systems have to hold up with non-technical users, several languages, limited connectivity and high-stakes institutional decisions.
| 21st → 2nd | 66,000 | 199 | 4,700+ |
|---|---|---|---|
| Himachal Pradesh's national rank in Grade 3, NAS 2021 → PARAKH 2024, while I built the state's data analysis and school-level reporting | Government teachers reached through state assessment, training and reporting systems | PM SHRI schools supported through digital workflows | School heads supported through leadership and training workflows |
- School-level analysis for PARAKH 2024. Regular data analysis and school-level reports that guided academic interventions ahead of the national learning-outcomes survey. The state also placed 3rd among states in Grades 6 and 9.
- SAARTHI, a teacher competency management system. AI-enabled, NPST-aligned assessment and professional growth: self and peer assessment, classroom observation, student feedback, teacher portfolios and AI-supported recommendations.
- Samarth and Manthan. A training and certification platform, and a mobile-first app for teacher diagnostics and field data capture.
- TLM Lab Portal and Lesson Plan Portal. Searchable, curriculum-linked teaching resources and lesson plans for teachers.
- AI systems: agent harnesses, tool use with approvals, multi-provider LLM routing, evaluation, RAG, OCR and document intelligence, multimodal AI
- Engineering: Python, TypeScript, React, Next.js, FastAPI, NestJS, Flutter, PostgreSQL, Docker, Cloudflare Workers, CI/CD
- Education systems: NPST, PM SHRI, NIPUN Bharat, PARAKH and NAS-aligned improvement, teacher professional development
I am open to collaborations in applied AI, LLM infrastructure and evaluation, and AI for education and public systems: [email protected]



