Mar 22, 2026 Semantic Retrieval Needs Confidence Gates: Evidence from a Labeled Sanity Benchmark Reasoning models did not break one filter by accident. The benchmark shows a broader pattern: semantic retrieval needs explicit confidence policies.
Mar 15, 2026 From Lexical Gates to Semantic Gating: What Our Sanity Retrieval Experiments Actually Show A stricter benchmark changed our interpretation: lexical overlap gating is weak, and robust retrieval needs semantic confidence policies.
Mar 8, 2026 Sanity Retrieval After Benchmarking: What Actually Holds Up in Production We re-ran our Sanity RAG setup with a labeled benchmark. Semantic retrieval wins on relevance, but only with confidence gating.
Mar 5, 2026 I built a compiler-first ML library for TypeScript builders PrisML is what happens when you decide ML should be a compile-time step, not another service to manage.
Mar 5, 2026 How I built PrisML: what’s actually implemented in the code today A technical behind-the-scenes of PrisML based on the real code: package architecture, train/check pipeline, ONNX runtime, and current MVP boundaries.
Mar 2, 2026 I built an AI bot to read posts and draft posts for me (sort of) The real problem wasn't writing. It was turning scattered reading into usable insights. So I built GRIST: harvest first, compose later.
Mar 1, 2026 I tested Sanity's search APIs on a real LLM pipeline. Not what I expected. I needed a retrieval layer for an LLM benchmark. Sanity had more built-in than I thought, and one feature I had not considered changed how I think about auditing.
Feb 10, 2026 Shipping Momentum Without Burnout A simple cadence for shipping weekly without losing the thread.
Nov 8, 2025 Designing Systems That Scale With Teams A few defaults that keep systems readable as headcount grows.