I build retrieval pipelines, concurrency-safe backends, and low-latency matching engines — with the benchmarks to back each one.
Three years of self-directed building alongside coursework — type-safe APIs, retrieval pipelines, and ML experiments that generalize past the toy dataset. My current obsession is the messy seam between deterministic systems and probabilistic models.
Flagships get the full case study — problem, numbers, stack. Everything else compresses into one scannable line.
A RAG documentation assistant that crawls a docs site, indexes it, and answers with grounded citations. Hybrid BM25 + dense retrieval, cross-encoder reranked, served sub-second by Groq Llama 3.3 70B.
High-concurrency ticketing built for flash sales. Atomic Redis/Lua seat locks give a contested seat exactly one winner; a FIFO waiting room streams queue position over Socket.IO.
A limit order book with a lock-free, zero-allocation C++20 matching core on strict price-time priority. A FastAPI gateway packs orders into 32-byte structs and pipes them over ZeroMQ.
CLI + library converting OpenAPI specs into Draw.io diagrams via a custom graph engine.
Distributed task scheduler — hand-built priority queue, worker pool, cron, live dashboard over WebSockets.
Fine-tuned VGG16 classifying MRI scans into four tumor classes, served via Flask.
Applied neural solvers to stiff singularly perturbed differential equations — the regime where classical numerical methods need punishing mesh refinement to resolve boundary layers.
Coursework across DSA, operating systems, DBMS, computer networks, and machine learning. Final-year focus on applied ML and full-stack systems.
4 completed programs, plus 10 Google Cloud Skills Boost badges — nothing hidden behind a click.