GitHub user musaabhasan added a comment to the discussion: [KPIP] Data Agent Engine — AI-Powered Autonomous Data Analysis for Kyuubi
The proposal is promising, but I would make the execution boundary very explicit before the engine becomes autonomous. A data agent that can inspect schemas, generate SQL, execute queries, and self-correct needs different controls from a stateless chat engine. A few areas seem important for the first implementation: - default to read-only execution, with write/DDL/DML disabled unless a deployment explicitly enables them; - enforce row limits, timeouts, cost limits, and query complexity limits before execution; - add a pre-execution SQL review layer that detects cross joins, unbounded scans, suspicious functions, and dialect-specific risky constructs; - preserve an evidence trail: user question, schema context used, generated SQL, execution engine, result sample, verification outcome, and final answer; - separate “answer confidence” from “query executed successfully”; a syntactically valid query can still answer the wrong business question. For evaluation, BIRD-style accuracy is useful, but production readiness also needs safety metrics: prevented unsafe queries, empty-result recovery, join-error recovery, hallucinated-column rate, dialect error rate, and whether the final answer cites the SQL/results that support it. The self-correction loop should also have a strict attempt budget. Otherwise the agent can turn one vague question into repeated expensive scans while trying to repair its own uncertainty. GitHub link: https://github.com/apache/kyuubi/discussions/7373#discussioncomment-16860857 ---- This is an automatically sent email for [email protected]. To unsubscribe, please send an email to: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
