About
I spent my first six years as an engineer where most of the hard problems live — behind the interface. Designing API architectures that serve hundreds of thousands of requests, building queue-based ingestion systems that compress 30-minute ETL jobs into 2-minute pipelines, and deploying production infrastructure on bare-metal VPS when managed services weren’t an option.
I don’t just write endpoints. I design the data flow, the failure modes, and the recovery paths.
At Northeastern, I turned that operational intuition toward ML and distributed real-time systems. I built NLP classifiers for clinical urgency detection on messy, imbalanced EHR data. I architected a server-authoritative multiplayer VR environment where terrain state synchronizes across Meta Quest headsets at frame-level precision. I designed a cross-chain analytics platform that ingests and indexes 17,000+ decentralized applications through concurrent Redis-backed pipelines.
The thread across all of it: I build systems that handle complexity at the infrastructure layer so the product layer can be simple. Whether that means a BullMQ pipeline with dead-letter recovery, a Netcode authority model that prevents client-side state drift, or a vector similarity engine that returns semantically relevant results in under 50ms.
Capabilities
Where most of my production work lives. API design, data modeling, authentication, queue-based processing, and service decomposition — owning the full backend from schema migrations to reverse proxy config.
Six years of React in production — from D3-powered analytics dashboards to enterprise micro-frontends. I build interfaces with the same rigor I apply to backend systems: typed, composable, maintainable.
Built during my MS and beyond — text classification, embedding-based retrieval, and semantic feedback systems. ML that actually ships: proper evaluation, class imbalance handling, and deployment paths that work in production.
Comfortable operating production systems without managed service dependencies — Docker, Nginx, CI/CD pipelines, and process management on bare-metal VPS. AWS-certified for when the cloud makes more sense.
How services decompose, where state authority lives, how failure propagates. Applied directly: VR server authority models, BullMQ job orchestration, CI/CD pipeline isolation, and service decomposition in enterprise platforms.
Multiplayer VR state synchronization at 72Hz across N Quest headsets, live leaderboards with WebSocket broadcasting, and terrain streaming over deterministic frame buffers. Low-latency as a hard constraint, not an afterthought.
Direction
End-to-end product engineering, from data model to UI
Service decomposition, consistency tradeoffs, scalable architecture
NLP, semantic systems, embeddings in production
API design, pipelines, infra that teams rely on
Off the clock
Hard support main (soft support in emergencies) — the one buying wards so the carry gets the glory. A decade in a game where 50 milliseconds decides a teamfight is probably where the real-time systems fixation started. The rank says noob; the hours say committed.
Eight years of school badminton and six of table tennis — CBSE Clusters and Nationals. These days it's badminton, table tennis, squash, and tennis. The competitive streak did not graduate.
FIFA / FC on whichever console is nearest, and any board game that ends friendships — Catan and Monopoly first among them.
Hamilton in F1. Federer in tennis, forever — with Alcaraz and Sinner making the post-Federer era watchable.
Iron Maiden, AC/DC, Queen, Linkin Park at full volume. Fred again.. and The Weeknd for range; J. Cole, Eminem, Joyner Lucas for the wordplay.
Succession, Breaking Bad, Game of Thrones on one shelf. Naruto, Death Note, Steins;Gate, Attack on Titan on the other — and Kuroko's Basketball counts as sports.
After Effects and Photoshop experiments, purely for fun. Genuinely not good at it. That is not the point.