Satellite imagery analysis doesn't always require massive data and compute. We show how to combine open data from OpenStreetMap and the UK Government's Renewable Energy Planning database with Tessera foundation model embeddings to map solar farms across the UK using a lightweight neural network.

OCaml represents less than 0.01% of training data for many large language models, creating an existential threat for the language community. This talk outlines three concrete steps the OCaml community can take to improve AI coding agent support and ensure the language's future relevance.

Can a model trained on satellite data really find brambles on the ground? We put a new model to the test with a field trip around Cambridge. The results were surprisingly good and taught us a lot about the model's strengths and weaknesses.

I stumbled onto terminal-bench a few weeks ago while researching datasets to evaluate agents. It contains around 120 tasks that need to be completed using a terminal..

A look at recent OCaml projects, from benchmarking AI code models and building new agentic tools to improvements in the garbage collector.

A simple MCP server that enables Claude Code (or any MCP host) to check things with gemini-cli

How well can locally-runnable language models handle OCaml code generation? We evaluate 19 open-weight LLMs on first-year Computer Science exercises, exploring the balance between model size, architecture, and reasoning capabilities for less mainstream programming languages.

Introducing opam-archive-dataset: a Parquet dataset containing code from OCaml packages, designed to improve performance of language models for OCaml development through better training data

Here's a workaround to get JSON output from Deepseek R1 and distills when using the llama.cpp OpenAI-compatible server endpoint

In this post we're going to discuss how to teach Bloom Filters new tricks. We'll start with examining Partitioned Bloom Filters and then look at ways we can generalise Bloom Filters to new and interesting uses. By the end of the post you will be able to use this generalised framework to come up with novel probabilistic data structures.

I was looking for some small Christmas stocking-fillers to give to techie friends and decided to try to find some interesting electronics boards from China.

In the end, I went with the WEMOS Lolin32 Lite which features Espressif's ESP32. If you're not familiar with the ESP32, it's an awesome little …