2+ day, 18+ hour ago   (1738+ words) I built a system that automatically discovers, verifies, and applies relevant requirements from earlier interactions without asking the user where they came from. I built a complete, working implementation in pure Python and shared actual benchmark numbers from real runs…...

1+ week, 6+ day ago   (1426+ words) What I learned after thinking more carefully about Structured Outputs on messy, incomplete data Three weeks after I turned on Structured Outputs for a pipeline that parsed payment confirmation messages into transaction records, I noticed that our reconciliation job started…...

4+ week, 3+ hour ago   (1680+ words) A one-shot pipeline commits to its first try: parse once, retrieve once, generate once, return whatever comes out. When retrieval comes back empty or the answer is half-formed, there is no second chance. A loop gives the pipeline one: notice…...

2+ week, 18+ hour ago   (1470+ words) How to apply the latest context engineering guidelines to your day-to-day data science work There are so many positive sides that come with using systems like Claude; all the repetitive, routine coding gets automated, researching is quicker, and debugging becomes…...

2+ week, 6+ day ago   (1842+ words) Enterprise Document Intelligence [Vol.1 #14C] - One hour with two people, six to ten fields, and the two signals that separate a real column from one that will break a filter later A lot of RAG work right now goes into letting…...

2+ week, 6+ day ago   (1470+ words) Enterprise Document Intelligence [Vol.1 #M3] – The ten positions the series argues from, and the map of every article that argues them This article is a manifesto of Enterprise Document Intelligence, a series that builds an enterprise RAG system from four bricks....

3+ week, 1+ hour ago   (1708+ words) A lightweight runtime layer that separates instructions, evidence, memory, and tool output before they reach the model This article is for anyone building agent systems who assemble prompts from multiple sources (retrieved documents, conversation history, tool outputs, or system instructions)…...

3+ week, 1+ day ago   (1761+ words) 28 debugging experiments reveal that AI struggles less with complexity than with missing information. If you’ve watched a coding agent in action, you’ve probably noticed the default workflow: let the AI handle the routine bugs, but step in when the problem…...

3+ week, 2+ day ago   (833+ words) LoRA fine-tuning solved our under-labeling problem. Whether it makes sense for you depends on three questions. Image classifiers can be built in many ways. The modern default approach is to run images through a third-party API which internally uses a…...

3+ week, 2+ day ago   (938+ words) A walkthrough of and the maths behind using low-capacity networks to acquire fine-grained scoring when only categorical labelling is available for training To be able to illustrate the work, I developed a toy example, which is a non-linear but deterministic…...