AI/ML Engineering Intern at A P Moller – Maersk, B.Tech CSE (AI/ML) from PES University. I build LLM systems for production and care most about the part everyone skips: checking the model's work: confidence scoring, human-review routing, and deterministic validation against hallucination.
- 💼 Maersk: AI document-extraction pipeline (GPT-4o vision + Azure Document Intelligence) running across 10+ countries, with Grafana observability and alerting
- 📄 Published at ICCIS 2025 (Springer LNCS): Temporal Belief Dynamics in BERT
- 🏆 Winner, HACK-AI 24-hour Data Analytics Hackathon
- 🌍 Merged open-source fixes in
pydantic-aiandREA
GPT-4o · Claude API · Gemini API · Hugging Face · Multi-agent systems · OCR · MCP · SSE · Claude Code
| Project | What it does | Stack |
|---|---|---|
| Provenance | AI-content detection beyond stylometry: watermarking, a paraphrase-robust retrieval ledger, and a live Attack Lab that red-teams every method | Python · TS |
| Groundtruth | Ranks Bengaluru's 369 wards by heat risk, recommends interventions within budget, and verifies every figure before showing it | Gemini · ADK · BigQuery |
| Il Colosseo (live) | Two Gemini agents debate any topic over three rounds with live web research, then an AI judge rules. Source-integrity gate, 10 Indian languages with voice | Gemini · FastAPI · SSE |
| DawaiSathi (live) | Type or photograph a medicine and get plain-language usage, side effects and warnings in 11 Indian languages, plus cheaper Jan Aushadhi generics | Claude · Next.js · Supabase |
| mcp-compensator | Ctrl+Z for AI agents: an MCP proxy that journals mutating tool calls and can undo them | Python · MCP |
| worth-a-ping | Reads your inbox and only interrupts you when it's worth it, with a reason why | Python |
| Aegis | Private recurring subscription payments on Starknet with publicly verifiable aggregates | TS · Starknet |
| Repo | Contribution | Status |
|---|---|---|
| pydantic/pydantic-ai | #10805 raise ContentFilterError when Azure blocks a Responses API prompt without content_filter_result (fix + regression test) |
|
| morluto/rea | #6137 fix reconstruction-ledger failure on whitespace-only graph labels (regression tests for both code paths) |



