RAG Evaluation Workbench
A practical toolkit for testing retrieval quality across datasets, prompts, and rerankers.
RAGevaluationagents
01
PROJECT
Systems I built
to understand systems.
These are working questions disguised as software. Each project begins with a problem I want to understand from first principles.
The Front Desk That Never Sleeps
A voice agent that remembers callers, identifies roofing emergencies, and hands off with context.
voice AImemoryhackathon
One-user RAG Coworker
A deliberately small email and document assistant built to make routing decisions observable.
LangGraphroutingRAG
How each project took shape.
Short English essays about the decisions, tests, and mistakes behind the software.
AGENT SYSTEMS
Designing an Intent Router From First Principles
A practical way to separate normal chat, retrieval, and tools without hiding uncertainty behind a large classifier.
RAG EVALUATION
Evaluating RAG Beyond Hit Rate
Retrieving a relevant chunk is only the beginning; a useful evaluation must follow the evidence into the final answer.
HACKATHON BUILD
Building the Front Desk That Never Sleeps
What a roofing voice agent taught me about memory, urgency, and designing a handoff that a real team can trust.