2026.08.14
⏱ 6 min read
Why Models Fail in Production: Lessons from 10,000 Satellite Rooftops
How high-resolution satellite imagery breaks standard object detectors, and why ensemble voting with IoU clustering beats hyperparameter tuning every time.
// DISPATCHES
Technical Blog & Essays
In-depth writing on Computer Vision, production RAG architectures, edge systems, and software reliability.
2026.06.22
⏱ 8 min read
RAG in the Trenches: Semantic Re-ranking, Chunking & Latency Budgets
Moving beyond naive top-k vector similarity: combining dense vector retrieval with cross-encoder re-ranking to cut repetitive course support queries by 30%.
2026.04.10
⏱ 7 min read
Zero-Video-Exit: Real-Time Edge Computer Vision on Constrained Hardware
Architecting in-store retail dwell time and foot-traffic heat maps on local edge hardware without a single raw video frame leaving the premises.
2026.02.05
⏱ 5 min read
What 15 Factory Floors Taught Me About Cloud SLAs and Engineering Reality
Before deep learning and LLMs, I programmed PLCs and SCADA systems in heavy industrial plants. Here is what industrial reliability teaches modern software engineers.
2025.11.18
⏱ 6 min read
Building High-Throughput LLM Gateways in Go: Token Buckets & Fallback Routing
Why we wrote our LLM reverse-proxy in Go with Redis token-bucket rate limiting to withstand provider outages and spike traffic.