@davidhahn

Prompt Engineering Lessons from Building a Problem Generator

21h ago · 4 min read · Prompt engineering only becomes interesting when you're prompting for structured, specific output that has to be reliable enough to feed downstream systems. A chatbot that gives a slightly different a

Structured Output — When to Use Prompting vs. Forced Tool Use

21h ago · 4 min read · Most LLM features eventually need the model to return structured data like a JSON object your application can parse and act on. Connecting LLM output to a database, a UI component, or a downstream API

The LLM-as-Judge Problem — Making Automated Evaluation Reliable

2d ago · 5 min read · Automated evaluation using an LLM sounds like an elegant solution until you understand its failure modes. The model playing the role of a teacher grading work has four well-documented ways to get it w

Designing an LLM System That Actually Solves a Real Problem

2d ago · 4 min read · Most LLM project ideas start from the technology. "What can I build with agents?" or "Let me try a RAG pipeline." That approach produces demos that are interesting for a day and abandoned by the weeke

Building RAG from Scratch — Embeddings, pgvector, and a Bug Worth Knowing

2d ago · 5 min read · RAG sounds complex until you break it into its actual steps: Query → embed query → search vector store → retrieve top N chunks → prompt + chunks → generate At its core, it's a retrieval problem with