Loading post…
AppleExplainer
Glyph: A Multi-Strategy Agentic System for Column Description and Sensitivity-Ontology Tagging of Enterprise Data Catalogs
Summary
Glyph addresses enterprise data lakes where tables accumulate faster than human stewards can document or classify them, leaving column descriptions and governance labels incomplete. It models column description generation and sensitivity-ontology tagging as cooperating LLM agents orchestrated as stateful graphs. The Descriptor uses active Retrieval-Augmented Generation to retrieve pipeline source code from an enterprise GitHub, while the Tagger runs description, line-of-business regex, and metadata strategies in parallel before combining ranked outputs with Reciprocal Rank Fusion. A fine-tuned six-layer MiniLM metadata encoder raised same-tag retrieval on an in-distribution held-out split from NDCG@10 0.55 to 0.92 and MAP@100 0.19 to 0.90, relative to the stock base encoder. Glyph also evaluates end-to-end multilabel tagging with a recall-weighted F2 objective across three groups and an ablation, emphasizing value-free, code-grounded operation, per-tag provenance, and graceful degradation for auditable production cataloging.
Context
Enterprise data lakes can accumulate undocumented and unclassified columns faster than human stewards can process them. The resulting documentation debt affects data discovery, access control, and regulatory compliance.
Approach / What changed
Glyph uses cooperating LLM agents in stateful graphs. The Descriptor retrieves pipeline source code on demand from an enterprise GitHub through a reasoning–acting tool loop, while the Tagger runs description, line-of-business regex, and metadata strategies in parallel. The metadata strategy uses a fine-tuned six-layer MiniLM contrastive encoder and vector database; Reciprocal Rank Fusion combines ranked outputs from all three strategies.
Takeaways
The Descriptor grounds column descriptions in the pipeline source code that produces each column, using active Retrieval-Augmented Generation rather than relying only on column values or metadata.
The Tagger combines three complementary strategies: description-based tagging, line-of-business regular expressions, and metadata retrieval from a fine-tuned contrastive encoder backed by a vector database.
Fine-tuning the six-layer MiniLM metadata encoder improved same-tag retrieval on an in-distribution held-out split from NDCG@10 0.55 to 0.92 and MAP@100 0.19 to 0.90.
FAQ
How does Glyph ground generated column descriptions?
The Descriptor retrieves the pipeline source code that produces each column from an enterprise GitHub. It uses a reasoning–acting tool loop for active Retrieval-Augmented Generation.
How does Glyph combine its tagging strategies?
The Tagger runs a description tagger, a line-of-business regex tagger, and a metadata tagger in parallel. It fuses their ranked outputs with Reciprocal Rank Fusion.
What improved the metadata tagger’s retrieval quality?
Glyph fine-tunes a six-layer MiniLM metadata encoder with an in-batch contrastive objective. On an in-distribution held-out split, same-tag retrieval increased from NDCG@10 0.55 to 0.92 and MAP@100 0.19 to 0.90 versus the stock base encoder.
Source and freshness
- Source
- machinelearning.apple.com
- Published
- Added to Yomu
Summaries are generated from the original by a language model. Check the original before relying on it.
Keep reading
Related reading
The Good Stuff
Paid picks we use ourselves. Links carry a referral code — a subscription through one earns us a commission at no extra cost to you.
The best engineering reads, once a week
One useful email. Unsubscribe anytime.