labrador: Evaluating a Multimodal Retriever - Part 2
This continues my previous post, which covered what labrador is and how it’s built: a local multimodal search over a Mac folder using EmbeddingGemma 2, with ...
Applied AI
Hands-on notes from building AI infrastructure: RAG pipelines, vector databases, taking LLM-backed services to production, and the formats and tooling that agents rely on. Each post is code-first and covers what worked and what didn’t.
Read oldest to newest, the posts follow a progression: RAG fundamentals, then vector store tradeoffs, then a production Q&A service, then knowledge formats for agents.
labrador: Local Multimodal Search on a Mac - Part 1
I have a folder problem. Somewhere on my laptop there is a receipt photo, a slide with the architecture diagram, a screenshot of an error with a request id i...
OKF: Google’s Markdown-Based Knowledge Format for AI Agents
Two days ago, Google Cloud published a spec called the Open Knowledge Format (OKF). It’s a v0.1 release — early, explicit about that — but the idea behind it...
Building a Document Q&A Service: From Prototype to Production
This is the post I wish I had six months ago when I was trying to take my first RAG prototype and turn it into something I could actually deploy and share wi...
Pinecone Goes Serverless: What Actually Changes
Pinecone announced their serverless architecture in early 2024, and after using both the pod-based and serverless versions on a few small projects, I have so...
Building Blocks of a RAG Pipeline
Over the past year I’ve watched “RAG” go from a niche research acronym to something that comes up in every engineering discussion involving LLMs. If you’ve b...