Mini-CDN
CDN simulation in Go: consistent hashing (or round-robin) across three edge caches, health-synced ring, load shedding (503 + X-Load-Shed), GET/HEAD failover, TTL/LRU disk cache, Prometheus/Grafana.
Technical write-ups for the systems I have built, plus links to other repos.
CDN simulation in Go: consistent hashing (or round-robin) across three edge caches, health-synced ring, load shedding (503 + X-Load-Shed), GET/HEAD failover, TTL/LRU disk cache, Prometheus/Grafana.
In-memory Tokio job queue with a bounded mpsc worker pool, exponential backoff with jitter, dead-letter list, and shutdown that waits until pending work is finished.
Queue-decoupled inference lab: FastAPI gateway, Redis Streams workers (mock or vLLM), XAUTOCLAIM recovery, bounded retries + DLQ, idempotency keys, 429 backpressure, lag/pending autoscaler.
Notes and demos for Linux isolation and networking: namespaces, cgroups, iptables/nftables, and a light eBPF pass (runs in a privileged Ubuntu container on macOS).
Fine-tuned Gemma-4-E4B on 3,000 Turkish RAG examples, training 0.23% of parameters in about 42 minutes on an A100, and published LoRA, 16-bit, and GGUF versions on Hugging Face.
Built a multi-agent RAG platform with a LangGraph supervisor, hybrid retrieval (dense + BM25 + cross-encoder reranking), and Qwen served on vLLM, deployed with Docker, Redis, Postgres, and JWT auth.
Built a local-first chatbot for grounded Turkish document Q&A with citations, supporting image, PDF, and voice input (faster-whisper STT, edge-tts TTS), web-search fallback, and LangSmith tracing.