FTI architecture, Uber's QueryGPT, Qdrant and a diagram trick
16 July 2025#
- Implement Data Collection + FTI (Feature, Training, Inference) architecture in any LLM app. Also building an LLM Twin — reading the LLM Engineer’s Handbook.
- Started reading The Intelligent Investor.
- Diagram trick: use an LLM to generate mermaid code, debug it in the live editor, then in draw.io use Arrange → Insert → Mermaid. This gives you a starter for complex flow diagrams. Very useful.
17 July 2025#
- An approach for natural language to SQL with huge tables: Uber’s QueryGPT.
18 July 2025#
- Used Qdrant Cloud for a vector DB. It has a free tier and is wonderful for building demos.