Backend engineer who takes LLMs out of the demo and into customers' real pipelines — messy data, real constraints, real consequences. I like owning the problem end to end: from API design to the sandbox it runs in to the developer who has to trust the output.
class Tanay:
role = "SDE-2 @ Appknox"
focus = ["agentic harnesses", "RAG + vector search", "LLM context engineering", "shipping AI into enterprise workflows"]
daily = ["Python", "Django", "LangGraph", "Qdrant", "Docker", "k8s"]
also = ["Rust (for fun)", "TypeScript", "Node.js"]
currently = "making agents earn trust: verify first, then fix"- AI that fits into how teams already work — LangGraph agents that build the right context for LLM reasoning, cut through noisy SAST/DAST signals, validate findings in sandboxes, and hand developers a fix they can merge, not a report to triage.
- Meeting developers where they are — MCP integration that puts remediation inside the IDE, and public REST APIs and async workers that enterprise clients wire into their own CI/CD.
- Making hard things usable — guided, sandboxed security testing for engineers who aren't security specialists.
- Scale problems — a mobile threat intelligence pipeline covering 3M+ apps across global stores: metadata feature engineering, logo similarity models, vector search.
| Project | What it is |
|---|---|
| farfetch 🦀 | Keyboard-driven TUI REST client in Rust. Git-branch-aware environments, cURL import, under 15 MB RAM. |
| anamnesis | Tiered agent memory for health tracking — daily logs get compressed into an index, then rolled up to long-term memory behind a human review checkpoint. MCP + FastAPI. |
| truenorth | Backtesting and paper-trading framework for NSE swing strategies, from bhavcopy ingestion to walk-forward validation. |
| text_summarisation | BART-based summarisation and topic extraction. |
Languages
Applied AI
Backend and data
Infra and DevOps
Security
Frontend (when needed)
