Case studies

Real systems — problem, architecture, decisions and results. No invented metrics.

Olive — Deliverable Tracking Platform

Shipped from first commit to production in under 2 weeks; still owned as the sole engineer in production, with an event-driven pipeline across 4 decoupled workers.

Read case study →

DISTRIBUTED SYSTEMS

Distributed Fault-Tolerant Task Queue

Stress-tested at 10,000 concurrent jobs with ~15,800 tasks/sec ingestion and 0% task loss under crash-recovery scenarios.

Read case study →

DISTRIBUTED SYSTEMS

Distributed Tiered Rate Limiter

Atomic Redis Lua enforcement across tiers — anonymous k6 runs hit 99.47% 429s with 12.37ms p95 latency; pro-tier allowance tracks theoretical refill.

Read case study →

EVENT-DRIVEN / MEDIA

StreamHub — Distributed Video Pipeline

Ingest and transcoding scale independently — clients PUT directly to object storage while a separate BullMQ transcoder builds an HLS ladder without blocking the API.

Read case study →

REAL-TIME SYSTEMS

LogStream

High-throughput log ingestion with Redis Streams consumer groups, SHA-256 deduplication, batch PostgreSQL writes, and live WebSocket metrics.

Read case study →

AI / SEARCH

Research Paper RAG System

3-stage CPU reranker (BM25 → RRF → Cross-Encoder) funnels retrieval 30 → 10 → 5 with ~400ms added latency — no GPU required.

Read case study →

AI / RAG

AI Medical Report Summarizer

Grounded medical-report summaries via LangChain RAG over Qdrant — answers stay tied to uploaded content rather than free-form generation.

Read case study →

REAL-TIME / COLLAB

DrawApp — Collaborative Whiteboard

Live collaboration scaled across Socket.IO instances via Redis Pub/Sub, with async PostgreSQL flush and email jobs off the draw path.

Read case study →

REAL-TIME / BACKEND

Real-Time Chess App

Sub-second multiplayer state via dedicated WebSockets, Redis for live games, and PostgreSQL for durable history with reconnection support.

Read case study →