Skip to content

feat(dla): M6 — business glossary + pattern catalog + feedback loop - #10

Merged
udaybhan05 merged 1 commit into
mainfrom
feat/m6-glossary-port
Jun 23, 2026
Merged

udaybhan05 merged 1 commit into
mainfrom
feat/m6-glossary-port

Conversation

@udaybhan05

Copy link
Copy Markdown
Collaborator

Brings M6 onto the new workspace layout (apps/dla).

What

  • Glossary (dla glossary build): extracts recurring name tokens, drafts a one-line definition per term via the LLM gateway, returns INSUFFICIENT_SIGNAL when usages don't support a confident definition; idempotent and SME-preserving.
  • Patterns (dla patterns detect): pure-Python detectors — star, snowflake, junction, audit-columns — over an in-memory schema graph (no DB connection).
  • Feedback loop: a confirmed glossary term (ai-drafted-edited/sme-authored) feeds the affected columns' describe grounding, so confirming a term re-drafts those descriptions on the next run (SME edits stay protected).
  • Schema: GlossaryEntryPayload, PatternPayload, PatternType; ThresholdsConfig gains glossary knobs.

Migration notes

  • Imports repointed to auropro_core.logging and auropro_llm.gateway; no new dependencies (openpyxl/rapidfuzz already present).

Tests

  • 138 apps/dla tests pass (unit + integration + eval); ruff + mypy clean. Deterministic evals for extractor accuracy and pattern detection; the LLM-definition-quality judge is model-gated.

Ported the M6 work onto the apps/dla workspace layout (imports repointed to
auropro-core logging and auropro-llm gateway).

- Glossary: recurring-token extractor + LLM definer (INSUFFICIENT_SIGNAL when
  unsure) + glossary_v1 prompt + dla glossary build; idempotent, SME-preserving
- Patterns: pure-Python detectors (star/snowflake/junction/audit) over an
  in-memory schema graph (no DB) + dla patterns detect
- Feedback loop: confirmed glossary terms feed a column's describe grounding,
  re-drafting columns that use the term
- Schema: GlossaryEntryPayload + PatternPayload + PatternType (reader dispatch);
  ThresholdsConfig glossary_min_recurrence + glossary_stop_tokens
- Tests: glossary extraction/definition/feedback, pattern detection (+ no-DB
  guard), deterministic glossary + pattern evals

138 apps/dla tests pass; ruff + mypy clean.
@udaybhan05
udaybhan05 merged commit f2d5812 into main Jun 23, 2026
0 of 2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant