Lab experiment

Local RAG

Exploring private, inspectable retrieval workflows for local technical knowledge.

Status
Exploring
Concepts
  • Retrieval
  • Local data
  • Evaluation

This diagram is a working model, not a deployed architecture.

A retrieval path under exploration: local source material is parsed and chunked, embedded into an index, queried for candidate passages, narrowed into a selected context window, and passed to a local model. An evaluation step feeds back into retrieval rather than being assumed correct.
01

next questions

Questions under investigation

  • How should local technical material be segmented and indexed?
  • How can retrieval quality be evaluated instead of assumed?
  • What context belongs in a prompt, and what should stay behind a tool boundary?
Previous experiment
AGENTCONTEXTTASKEXECUTEVERIFYRETRY
Agentic Development

Still exploring

Thinking about the same problem?

I'm always interested in what works, what fails, and how to test the difference.

Next experiment
PROTOCOL BOUNDARYCLIENTMODELTOOLSRESOURCESSERVERSREQUEST
MCP & Tooling