All Posts

Thoughts on building AI systems

Reverse-engineering my car with GLM-5.3-Flash

My car died 500 meters from home. No open-source OBD software could talk to it, so I reverse-engineered the connection with GLM-5.3-Flash: protocol probing on a raw serial port, with a pure-C ELM327 reader.

Neuroevolution will thrive when the bubble pop

A case for neuroevolution after the generative-AI hype cycle. Why evolved structure, open-ended search, and hybrid learning systems may outlast the current benchmark chase—and where they are most likely to win.

Lessons Learned from Building Agents

Practical insights from developing AgenticSeek and production agent systems at CNRS. Covering provider resilience, caching strategies, failure modes, and the reality of current MCP ecosystems. A pragmatic guide for engineers building in the agent space.

Using Python to classify sounds (with PyTorch)

Everything you need to know about audio classification using deep learning. From data collection on YouTube to spectrogram conversion and CNN implementation with PyTorch — a practical guide for Python developers getting started with sound classification.

Topics

AI Agents LLM Systems Engineering MLOps Local AI Research