Projects

Distributed Job Queue

SQS-style queue on Java 21 and Spring Boot with visibility timeouts, exponential-backoff retries, a dead-letter queue, and Prometheus/Grafana observability.

Java · Spring Boot · PostgreSQL · Prometheus

GPU Cluster Scheduler

Kubernetes-inspired scheduler that places ML training and inference jobs across a heterogeneous GPU fleet, with leader election, node agents over gRPC, and a live cluster dashboard.

Go · gRPC · Kubernetes · Next.js

Distributed Search Engine

Full-text search built from scratch: sharded inverted indexes in C++, a Go coordinator that fans out and merges BM25-ranked results, Redis caching, and an observability dashboard.

C++ · Go · gRPC · TypeScript

ML Ads Ranking System

A Python pipeline trains a LightGBM CTR model; a Go service walks the tree ensemble natively at ~356 ns/op with zero allocations, matching Python predictions to 1e-9.

Go · Python · LightGBM · Redis

LLM Semantic Caching Layer

OpenAI-compatible proxy that answers semantically similar prompts from a Redis vector index — 73.8% lower cost and 99.3% lower P50 latency on a simulated 2,000-request workload.

Python · Redis Stack · Embeddings · Vector Search

gRPC Microservice

C++ order-management service over gRPC and Protocol Buffers: unary and server-streaming RPCs, a logging interceptor, thread-safe store, and graceful shutdown.

C++ · gRPC · Protobuf · CMake