Why Small Local Models Matter for Enterprise Automation
Small local models can handle recurring, bounded tasks where cost, privacy, latency, and operational control matter more than general-purpose brilliance.
Research blog
Applied AI field notes.
Seeded essays on model evaluation, RAG, prompt routing, local infrastructure, and the bridge from classical computer vision to modern AI.
Qwen Benchmarking: Lessons from a Full Factorial Experiment
A preliminary look at model size, quantisation, context settings, and inference behavior as interacting variables rather than isolated knobs.
Designing a Prompt Router for Enterprise AI
Prompt routing turns AI adoption into architecture: classify the request, select the right model, enforce policy, and preserve auditability.
RAG Systems: From Embeddings to Retrieval-Augmented Answers
Practical RAG depends on retrieval quality, lexical search, vector search, reranking, and answer constraints, not just embeddings.
Local AI Infrastructure: Balancing Cost, VRAM, Context, and Throughput
A field note on LM Studio, Ollama, llama.cpp, vLLM exploration, and the tradeoffs of running inference locally.
From Classical Computer Vision to Modern Multimodal AI
A line from 2012 HAAR and QR experiments to modern multimodal systems: the goal remains turning perception into useful action.