Projects

image_description — a Python toolkit for my photography workflow: it generates descriptions and alt text for images, publishes them to Mastodon, Pixelfed, Tumblr and Bluesky, tracks engagement, and promotes the best photos to this site’s gallery.

The galet stack

A set of provider-neutral Python packages for building LLM agent applications — split by responsibility so each piece stays usable on its own:

galet — provider-agnostic LLM, embedding and image-generation layer: chat and tool calling across OpenAI, DeepSeek, Gemini, Mistral and Ollama, declarative model routing, and a model catalog that resolves by capability.

galet-memory — the memory layer for agent applications: episodic, semantic, procedural and working memory contracts, with reusable SQLite and vector-backed persistence and retrieval.

galet-prompt-builder — prompt compilation: it assembles bounded prompts from instructions, memory and retrieval candidates, applying per-section token budgets, relevance selection and deduplication.

galet-tools — an MCP server exposing the galet stack as tools.

This site — a Jekyll/GitHub Pages site I publish to from the command line, with an automated photo gallery that ranks and thumbnails itself.