Systems I've actually built.

An enterprise retrieval-augmented-generation system over technical PDF manuals — four LLM providers (Phi-4 and Llama local; Mistral and GPT cloud), built on Clean Architecture.

• 4 language models, 5 embedding strategies, 4 chunking algorithms, 4 retrieval strategies
• Sub-50ms vector search on ChromaDB; under 8s end-to-end query time
• Enterprise resilience patterns: circuit breaker, retry with backoff, timeout protection
• 66/66 tests passing with 100% coverage on critical paths

A real-time, multi-stage data-processing pipeline that demonstrates queue management, backpressure, and automatic load balancing — with a live dashboard.

• 3-stage pipeline (Generator → Processor → Aggregator) with configurable worker pools
• Automatic backpressure detection when queues cross thresholds
• Circuit breaker pattern with health state machine (HEALTHY → UNHEALTHY → DEGRADED → DEAD → RECOVERED)
• Real-time SignalR dashboard with zero-polling updates (<2s latency)

REST service registration and health monitoring platform with approval workflows, real-time visibility, and pluggable storage backends.

• Health-state model with automatic recovery logic
• Heartbeat pipeline tuned for low-latency health checks
• SignalR dashboard for registration and approvals
• Hardened API surface with rate limits and security headers

Home Assistant weather visualization card with interactive charts, unit conversion, timezone support, and HACS distribution.

• Daily / hourly forecast toggle with runtime switching
• 6-level comfort-zone temperature gradient (cold blue → hot red)
• Smart unit conversion: °F ↔ °C, inHg ↔ mmHg ↔ hPa, mph ↔ km/h
• 23 languages, animated icons, customizable layouts