Acquiring signal…
Backend & systems engineer. I build software that holds up under load, from a C++ TCP server written on raw sockets to ML that ships, and I write the automated test harness that proves it on every run.
I work at the layers most peopleabstract away — and I prove whatI build actually works.
CS + Data Science at SRM Institute, currently an SDET Intern at HealthEdge. I write production code in C++, Node.js, and Python — from a hand-rolled TCP server with an epoll event loop to multi-tenant Postgres platforms — train and explain ML models with XGBoost and SHAP, and automate the regression that keeps it all honest. The thread across everything: numbers I can point to, not adjectives.
Built an in-memory key-value server from scratch in C++ with an epoll event loop and full RESP protocol parsing — no framework, just sockets. Sustains 150K+ commands/sec (1M+ pipelined) with pub/sub messaging.
Shipped a five-hazard disaster-prediction platform across 15 Indian regions. XGBoost models made auditable with SHAP explainability, served sub-second through FastAPI on live NASA and weather feeds.
Built real-time collaboration handling 50+ concurrent WebSocket connections at sub-100ms delivery — Socket.IO messaging, JWT auth and RBAC across 60+ REST endpoints.
Trained a classifier on 70K+ WELFake articles to ~94% accuracy with TF-IDF and logistic regression, plus live URL content extraction. Deployed and public.
Throughput and latency are numbers I benchmark and reproduce on demand, never adjectives on a slide.
I automate regression and API contracts with Playwright, Katalon and Postman, so "it works" is something I prove on every run — not something I hope for.
Reuse what already exists, prefer dynamic waits over brittle delays, and keep every change minimal and reviewable.