Software engineer working across the stack — from sub-millisecond C++ trading gateways and TLS/ACME certificate infrastructure to MCP multi-agent orchestration and production React. I care about correctness, latency, and code that ships.
SS// add public/profile.jpgI like problems where correctness and latency both matter — and I've built for all three sides of the stack.
Most recently I shipped production code at UKG on a multi-tenant recruiting SaaS, before that I engineered a C++ crypto-exchange gateway at Go Quant handling 750+ trading pairs at 0.5 ms per message. Alongside internships I build systems from specs — a TLS/ACME certificate service straight from RFC 8555, and a multi-agent MCP orchestrator with 391 tests.
I'm finishing my IMSc in Mathematics & Computing at BIT Mesra and I'm looking for roles where I can go deep on infrastructure, AI systems, or product engineering.
Low-latency, protocol-level engineering — C++ WebSocket gateways, order-management systems, TLS/ACME, and high-throughput data codecs.
Multi-agent orchestration over MCP, RAG & LangChain pipelines, and applied deep learning — from LSTM fleet optimization to task-DAG planners.
Production React & TypeScript front-ends, Node & ASP.NET services, Postgres/Mongo data layers — with accessibility and testing built in.
Four internships across enterprise SaaS, high-frequency trading, and industrial AI.
Systems built for correctness — implemented straight from RFCs and protocols, with real test coverage.
A TLS certificate lifecycle service implementing ACME (RFC 8555) from the specification — issuing real certificates through JWS-signed requests and HTTP-01 domain validation against Let's Encrypt's Pebble CA.
A two-pass TLS probe reads expired and self-signed certificates, verifies handshake rejects, and feeds a deterministic X.509 risk engine that scores expiry urgency and hygiene independently.
A multi-agent orchestrator that compiles natural-language goals into a Zod-validated task DAG, dispatching topological waves concurrently to capability-scoped MCP specialists with runtime-enforced tool ownership.
Bounded replanning runs over an append-only event log with derived state, so JSONL replay reuses the live execution path — scored across 15 golden scenarios.
An explainable acquisition-fit layer for SaaSquatch Leads — scores a company 0–100 as a search-fund acquisition target, not a sales lead, across six weighted factors: succession pressure, buy-box fit, digital-modernization upside, niche fragmentation, contactability, and liveness.
Every score shows its work — subscore, weight, evidence sentence, and source link per factor — plus a thesis-coverage figure so a 78 backed by 40% of evidence reads differently from one backed by 95%. The engine is deterministic with no LLM in the scoring path, and a fixture keeps the Python and TypeScript implementations provably in agreement.
Turns a resume into a published developer portfolio in one click — a guided wizard walks through 12 templates and publishes a public page at /p/<username>, backed by Next.js 16 and Supabase.
Sign-in is fully passwordless — magic links via Supabase OTP, with both PKCE and default email-template link formats handled and failures routed to a visible error state instead of a silent bounce. An npm run doctor script catches a bad env, a deleted Supabase project, or a missing schema before it reaches a user.
An AI-powered wardrobe assistant — upload or drag-and-drop clothes into a virtual closet, pick a style, and GPT-4o generates ranked outfit combinations with a score and a short rationale for each.
React and SASS on the front end, a Node/Express API in between, and IndexedDB persisting images and outfit history entirely client-side — no server-side storage of anyone's wardrobe.
AI interview platform that asks questions, evaluates answers, generates follow-ups, and scores fluency and grammar.
Competitive-programming platform — Next.js and a Monaco editor, Judge0-powered execution, a leaderboard, and a background judging worker.
Async approval workflow for leave and reimbursement requests — Flutter front end, Firebase/Firestore for real-time status across requester, approver and admin roles.
Explore the rest of my open-source work, experiments and course projects on GitHub.
I'm open to SDE and AI-systems roles, and always happy to talk about low-latency systems, agents, or anything technical. My inbox is open.