software engineer · pune, india
i build reliable AI systems, developer tools, and products that hold up after the demo.
been here since 24.000000000 years.
i work across retrieval, agent infrastructure, evaluation, security, and the ordinary software engineering that makes those systems useful in production. today, i am a software engineer at HSBC.
lately: building Redpen, shipping AgentCI, and contributing reliability, search, and developer-experience improvements to open-source AI infrastructure.
live signal
rolling 12 months · refreshed daily from public GitHub activity
selected work
a small set of things i can explain end to end: why they exist, where they fail, and what makes them more than a demo.
“done” is a claim. evidence makes it true.
a CLI that checks coding-agent completion claims against fresh repository evidence. it imports Codex claims as untrusted input, runs applicable tests and builds, then reports each claim as proven, failed, or unverified.
TypeScript · CLI · Git · npm
CI/CD for the behaviour of AI agents.
a release gate that compares candidate agents against production and blocks regressions in correctness, groundedness, permissions, safety, latency, and cost. every decision is backed by visible retrieval traces.
Next.js · TypeScript · OpenAI · BM25 · Vitest
local-first applied AI · 2026
reviewed, private translation for field learning material.
a CPU-first workspace for translating English, Hindi, Marathi, and Odia documents and media locally. humans review every result before it becomes an integrity-checked package that works offline in the field.
Python · FastAPI · local models · FFmpeg · Windows
AI-native phishing simulation for small teams: realistic campaigns, risk visibility, and training before a real attacker arrives.
turns arXiv papers into layered explanations, concept maps, quizzes, and visual learning paths, with structured-output fallbacks.
a vector database built from scratch with persistence, multiple distance metrics, document chunking, and exact, HNSW, IVF, and KD-tree indexes.
open source
merged work across failure handling, observability, search, security, and interfaces other developers depend on.
shipped CLI JSON and watch modes, doctor checks, nested-project resolution, lifecycle routing fixes, dashboard failure surfacing, observability bundles, merge-conflict links, and project-scoped terminal resolution.
inspect the merged work ↗added organization-scoped API-key validation, model configuration for prompt versions, strict template-variable checks, hybrid registry search, verified self-updates, and CLI file opening.
inspect the merged work ↗added interpretable memory-search score explanations and carried that capability through the REST server, so retrieval results can expose why they ranked where they did.
inspect the merged work ↗production work
build and maintain Java/Spring systems that enrich real-time corporate-action events across vendor feeds, XML/JSON transformations, and ActiveMQ pipelines. also built an internal RAG assistant covering ingestion, embeddings, retrieval, context assembly, and LLM-backed answers.
built commodity report-release analytics over historical market data and automated Eikon/TRTH extraction, cleaning, and research datasets; the resulting strategy work improved measured success rate by 15%.
how i work
make behaviour measurable. evals, traces, regression suites, and explicit failure states beat a confident demo.
own the whole path. interface, API, data, deployment, documentation, and the awkward operational edges in between.
write things down. clear architecture, trade-offs, runbooks, and limits make software easier to trust and improve.
contact
i am open to serious AI engineering, applied AI, and agent-infrastructure roles, especially where reliability, product judgment, and hands-on building all matter.