Backend engineer — I design and operate distributed systems in production. Java & Spring at scale, Go for systems projects, exploring LLMs & AI infrastructure.
Interactive case studies of systems I designed and shipped at Octro. Trace a request through the architecture, break it, and watch how it recovers — with the tradeoff behind every decision:
| Case study | What it shows | Stack |
|---|---|---|
| Async Notification Pipeline → | Decoupling push delivery from the request path so a slow third party can't stall the app; scaling it to 100M users | Amazon SQS · Redis · DLQ |
| Account Migration & Merge → | Merging guest → registered accounts — safe under partial failure, idempotent under retries, auditable end-to-end | 3 services · MongoDB · Redis |
| Real-time Leaderboard → | Live ranks with O(log N) updates, one push per interval, and recovery when the cache dies | Redis ZSET · WebSockets |
▶ Open the interactive case studies →
Backend engineer at Octro Inc (Best Performer), building distributed Java/Spring microservices across a 54-service platform on AWS — WebSockets, RabbitMQ, Redis, MongoDB, ZooKeeper, and async Amazon SQS pipelines, plus a polyglot Java ↔ TypeScript Lambda compute layer. I ship features end-to-end and carry on-call.
Self-authored side projects — real code with tests, metrics, and design docs. Grouped by theme.
Distributed systems
| Project | What it demonstrates | Stack |
|---|---|---|
| wal-kv-store | Durable key-value store: write-ahead log → crash recovery → compaction → Raft replication (3-node), with fault-injection crash tests + a design doc | Go · hashicorp/raft |
| code-execution-engine | Runs untrusted code in hardened Docker sandboxes (drop-caps · no network · PID/mem limits), Redis job queue + worker pool, SSE result streaming, Prometheus metrics + threat model | Go · Docker · Redis |
| distributed-rate-limiter | Pluggable strategies (token-bucket / sliding-window / Redis+Lua for cross-instance limits), HTTP + gRPC APIs, Prometheus, Docker Compose | Go · Redis · gRPC |
| payments-ledger | Double-entry accounting service: idempotent transfers + transactional outbox — no double-spend, no dual-write races | Go · Postgres |
| cdc-pipeline | Change-data-capture: DB change stream → Kafka → idempotent sink, with a dead-letter path + replication-lag metrics | Go · Mongo · Kafka |
AI infrastructure & products
| Project | What it demonstrates | Stack |
|---|---|---|
| llm-gateway | OpenAI-compatible gateway that routes each request to the cheapest capable model + a semantic response cache + per-key rate limiting + cost metrics | Go · Redis |
| vidbite | AI YouTube-video summarizer — full-stack, JWT auth, Dockerized and deploy-ready | React · Flask · Gemini |
| livepolls | Real-time audience polling with a live word cloud (Slido-style) over WebSockets | socket.io · Redis pub/sub |
Also: go-load-balancer (HTTP LB · health checks · hot config reload via fsnotify) · EnvBox (env manager with client-side AES-GCM encryption) · ssh-server · go-cli-tool
Languages
Backend & Infrastructure
