BLOG
AI engineering notes
Articles on what I build day to day: LLM agents in the dev cycle, computer vision at the edge and digital twins over critical infrastructure. No hype, just engineering.
GEO: How to Get ChatGPT and Perplexity to Cite Your Site
Generative engine optimization (GEO): llms.txt, schema.org (Person, FAQPage, hasCredential), extractable content and E-E-A-T so AI cites your website.
Real-time IoT dashboard: from sensor to decision
How to build a real-time IoT dashboard: MQTT/P2P ingestion, time-series storage, live visualization, smart alerts, and turning raw data into a useful decision.
Resilient, free AI: LLM fallback that never goes down
How I built a free LLM fallback chain (Groq, OpenRouter, Pollinations) with graceful degradation: chained free tiers that survive quota limits and timeouts.
Fast Websites with Astro: Islands, Zero JS and Core Web Vitals
How I build fast websites with Astro: islands architecture, zero JS by default, CSS purging and Core Web Vitals (LCP/INP) to improve SEO and conversion.
Claude Code and MCP: LLM agents in your development loop
How I run LLM agents in production with Claude Code and the Model Context Protocol (MCP): context engineering, tools, and the limits worth setting.
On-device AI on mobile: NPU, Whisper and real privacy
How to run on-device AI on mobile with TFLite, NNAPI and the GPU delegate, transcribe voice with local Whisper, and why on-device privacy isn't marketing.
From idea to MVP with AI in weeks, without burning your budget
How I validate and ship a fast, cheap MVP with AI: modern stack, cost-controlled AI, a phased process and clear deliverables. For founders in a hurry.
Practical RAG that actually works: a field guide
Practical RAG beyond the demo: chunking, local embeddings with FastEmbed/ONNX, hybrid retrieval, reranking, retrieval evaluation and stopping hallucinations.
LLM agents in production: from prototype to reliable system
Practical guide to LLM agents in production: orchestration, function-calling, memory, guardrails, evaluation, cost and latency. What breaks beyond the demo.
Computer vision with YOLOv8 at the edge
Field notes on shipping on-device object detection with YOLOv8 at the edge: latency, false positives, and why preprocessing matters more than the model.
Digital twins with BIM and GIS: from 3D viewer to production
What it really takes to ship a BIM + GIS digital twin to production: 3D viewer performance, live data, and the jump from pretty model to a usable tool.