mba/chunking_strategies_evaluation

Go to file

Mahdi Bazrafshan d5ccec2f2c feat(config): apply fixed_size ±3 as stabilized defaults

Why:
- Final benchmark selected fixed_size with neighbor expansion ±3 as the winning Candidate.

Changes:
- Set neighbor_prev/neighbor_next defaults to 3 in Settings.
- Default Query tab to fixed_size ±3/3 and Benchmarks to single-strategy fixed_size.

Impact:
- API and Dashboard runs use ±3 unless explicitly overridden.

Co-authored-by: Cursor <cursoragent@cursor.com>

2026-08-22 13:27:15 +03:30

docs: capture pdf domain language and deferred ingestion backlog

2026-08-02 16:59:42 +03:30

docs: record fixed_size as the winning strategy family

2026-08-17 14:22:38 +03:30

docs(data): add 21-question evaluation dataset for benchmarking

2026-07-26 13:48:36 +03:30

chore(scripts): add neighbor expansion benchmark sweep

2026-08-10 14:13:07 +03:30

feat(config): apply fixed_size ±3 as stabilized defaults

2026-08-22 13:27:15 +03:30

.env.example

feat(chunking): add embedding model registry and ollama config

2026-08-10 14:12:36 +03:30

.gitignore

Initial commit

2026-07-21 11:27:49 +03:30

CONTEXT.md

docs: record decision board language and adr-0026

2026-08-17 14:22:28 +03:30

LICENSE

Initial commit

2026-07-21 11:27:49 +03:30

pyproject.toml

build(deps): add pymupdf for text pdf extraction

2026-08-02 16:59:42 +03:30

README.md

Initial commit

2026-07-21 11:27:49 +03:30

README.md

chunking_strategies_evaluation

This repo is in order to reach multiple chunking strategies and evaulate them