field notes · zero fluff

writings

what i learn, as i learn it — evals, agents, rag, reasoning. written for people who ship.

11 notes

01 may 2026 three flavors of simulating an agent pipeline

02 may 2026 evolving an eval harness from one bash script to a merge gate

03 oct 2025 chunk twice, retrieve once: double-pass semantic chunking

04 sep 2025 building a context system that agents can actually use

05 jul 2025 teaching a small llm to filter its own context

06 jun 2025 grpo + lora: making a 3b model reason

07 jun 2025 colqwenrag: rag without ocr, for legal papers

08 jun 2025 fine-tuning embeddings until legal-rag stops missing

09 mar 2025 rolling your own reasoner: grpo + sft from scratch

10 mar 2025 how to make a model think: the deepseek recipe

11 mar 2025 a llm that drafts indian legal contracts