Project record

DocNeedle

Local-first SQLite FTS5 tool for indexing docs, retrieving chunks, and checking stale terminology.

Iterating State since May 29, 2026 Python · SQLite · FTS5
Architectural shelving geometry representing indexed documentation
Local retrieval — shelves as navigable document space.

Current state. Public README and v0.2.0 tag document a focused CLI surface while keeping benchmark harnesses explicitly separate from normal user workflows.

Context

DocNeedle is the most portfolio-ready repository in the current evidence set because the README clearly explains what the tool does today and what is still author-environment infrastructure.

Problem

Developer tools often blur stable commands, internal experiments, and benchmarking machinery into one narrative. That forces readers to guess what is actually safe to use.

Constraints

This case study does not publish benchmark numbers or treat hidden scripts as productized features. It keeps the story close to the documented CLI.

Approach

The public workflow is deliberately small: index a markdown corpus, search ranked chunks, read a chunk, and check for stale or conflicting terminology. The README also labels benchmark scripts as repo-specific and asks readers to patch hardcoded paths before reuse.

Evidence

Public evidence includes the README, the repository tree, and the v0.2.0 tag. That is enough for versioned and open-source badges without inflating performance claims.

Decisions

The case study presents the documented separation directly without inventing an ADR record.

Outcome / current state

DocNeedle is iterating. It already supports a strong public case study because the product surface and the evaluation harness are separated clearly enough to inspect.

Next

If public benchmark reports are added later, the case study can expand with attributable results while keeping the same core story.