Best Machine Learning Subagents

AI-powered Personal Knowledge Assistance in a folder. Built on the ICOR methodology. Plain markdown. Any LLM. Yours forever.

Deterministic (not LLM). Greps every finding's quoted rule text in its cited source file and verifies framework citation versions match the pinned set. Mismatches → quarantine.

Dispatch brief for the citation-verifier step — runs the deterministic, non-LLM grep gate (tools/pci-sss/citation_verify.py) over every PCI SSS v2.0 verdict's cited evidence and quarantines any that cannot be confirmed. The verifier is code, not judgement; this brief governs how an agent invokes it

Analyze the local source repo, detect the AI framework and LLM SDK usage, map all call sites, identify prompts, and enumerate user-visible behavior-deltas for the Bedrock migration. Returns a structured analysis object.

Parse LLM API logs from the local repo, extract prompt/response pairs, and build a golden dataset (prompts.jsonl) for evaluation. Returns a structured ingestion object.

Run each golden prompt against the target Bedrock model via the pinned uv harness, score with LLM-as-judge, and report a pass rate. Handles throttling with backoff; returns a structured eval object, or a partial/blocked control state.

Execute DataHub search, browse, and lineage operations, retrieve entity metadata, and return structured results. Used by the datahub-search and datahub-lineage skills to delegate catalog queries. <example> Context: User wants to find all Snowflake datasets with PII tags. user: "Search DataHub for Sn

Rewrite LLM SDK calls to Amazon Bedrock on a dedicated git branch, swap dependencies, generate tests, and apply the user-confirmed behavior-delta decisions. Returns a structured rewrite object.

Designs Pixeltable schemas — tables, views/iterators, computed-column chains, embedding indexes, and UDFs — for multimodal and ML data pipelines. Use when the user needs to model a data/AI workflow or decide between a view, a computed column, and a UDF.