Projects
Filter by tag. All writeups include what broke, eval numbers where they exist, and the honest limitations.
-
aegis-rag
active · 2026-04A modular RAG framework for benchmarking retrieval strategies — vector vs. keyword, with and without rerankers — before anything touches an LLM.
- AI/ML
-
Local LLM orchestration and benchmarking
active · 2026-04Running Qwen and Llama variants on 16GB consumer hardware via llama.cpp — measuring the real trade-off between tokens per second and logic quality on quantized GGUF models.
Qwen 4B at ~50 TPS on 16GB · 8B hits the swap wall at default quants
- AI/ML
- Systems
-
EU AI Footprint Scanner
active · 2026-04AST-based static analysis that detects AI/ML library use across a Python codebase, classified into simplified EU AI Act risk tiers. The first product I'm shipping under Argus Intelligence.
- AI/ML
- Tools
-
Edge routing proxy
shipped · 2026-04A multi-layer reverse proxy for upstream-resilient routing — OpenResty, Couchbase, and a TypeScript control plane. Serves ~10M requests/day in production from a 2 vCPU / 2 GB node.
10M req/day in production · 2 vCPU GCP e2-small · 70k concurrent at 100% in load test
- Backend
- Systems
-
Rotation service
active · 2026-04A stateless fleet health + load-balancing service for the proxy platform — weighted random selection with latency-based weights, hysteresis-protected health states, CAS-safe updates against Couchbase.
- Backend
- Systems
-
YOLO custom object detection
exploratory · 2026-04Transfer-learning YOLO on small, custom datasets for objects not in COCO — working the dataset quality and overfitting problems more than the model architecture.
- AI/ML