Blog
AI Engineering Deep-Dives
Technical notes on LLM inference, RAG pipelines, agent reliability, and AI product engineering.
Compiling YOLOv11-Nano and llama.cpp on Android NDK for Sub-30ms Field Inference
Cross-compiling vision and language models for ARM64 with CMake, building a JNI bridge with UTF-8 Devanagari handling, and running inference on $100 phones under 1.5GB RAM.
Deterministic ETL Safety Gates: Restricting Agrochemicals via Embedded SQLite FTS5
How hard-coded Economic Threshold Levels and BM25-ranked ICAR advisories enforce bio-first prescriptions on-device. Chemical dosages locked to a verified registry.
Agent Reliability: Building Tool-Call Safety Gates for Production AI Agents
How we built Jev — a three-signal safety gate for AI agent tool calls. Risk classification, intent relevance scoring, and prompt injection detection in under 200ms.
Building Production RAG: From Document Ingestion to Cited Answers at 50K Queries/Day
How we built the RAG pipeline behind Wokay's Woby AI assistant — document chunking strategies, hybrid retrieval, and citation-grounded generation at enterprise scale.
Cascading Model Routing: How We Cut LLM Inference Costs 85% Without Losing Quality
A practical guide to building a cascading model router — Haiku for simple tasks, Sonnet for reasoning, Opus for complex analysis — achieving 85%+ blended margins.
Meta to Charge for WhatsApp Business Messages — What It Means and How to Cut Costs
Starting October 2026, Meta will charge for every WhatsApp Business API reply. We break down the timeline, the cost impact, and 6 strategies to reduce the bill.