DATA OPS
BigQuery Regression LLM Root-Cause Explainer
On a detected cost spike it sends the old and new query SQL plus job stats to an LLM, which explains in plain English why the query got more expensive and suggests a concrete fix.
How it runs
The automated pipeline, trigger to output.
- TriggerDaily schedule
- ActionFind largest regressor + job statsBigQuery
- ActionPull current and previous SQLGitHub
- ActionLLM explains root cause + suggests fixOpenAI
- OutputSend diagnosis to SlackSlack
What it does
Turns a raw slot-hour spike into a human-readable root-cause: it feeds the before/after SQL and BigQuery job statistics to an LLM that explains the regression (e.g. a dropped partition filter, a new cross join) and proposes a fix.
When to use it
When your team can detect cost spikes but loses time diagnosing *why* a query got slower. Use it to get a first-pass diagnosis attached to every regression alert.
How it works
- 1A scheduled trigger fires daily.
- 2A BigQuery query identifies the scheduled query with the largest slot-hour increase versus baseline, along with bytes scanned and stage timing.
- 3A GitHub action pulls the current and previous SQL for that query.
- 4An OpenAI step receives both SQL versions and the job stats and returns a root-cause explanation plus a suggested optimization.
- 5A Slack message delivers the spike metrics, the LLM diagnosis, and the proposed fix.
Set it up
What you configure once, before turning it on.
- 1Connect BigQueryDatasets, queries, schemas.
- 2Connect GitHubRepos, issues, pull requests, actions.
- 3Connect OpenAIModels, embeddings, files.
- 4Connect SlackChannels, DMs, threads, mentions.
- 5Set each agent's modelWe leave models unset so you pick the tier — fast + cheap, or top-quality.
- 6Tune it to your dataEdit the prompts, filters, and field mappings so it matches how your team works.
- 7Test, then turn it onRun once against a sample, confirm the output, then enable the trigger.
More Data Ops workflows
Snowflake column type-drift sentinel with Linear fix ticket
Snapshots the data types of every column in your tracked Snowflake schemas on a schedule, diffs against the last snapshot.
Daily BigQuery Scheduled-Query Cost Attribution to Owners
Each morning, totals the prior day's on-demand bytes-billed per scheduled query, maps each query to its owner from a label, and posts a per-owner cost leaderboard to Slack.
BigQuery dropped/renamed column sentinel with PagerDuty incident
Detects when a column is dropped or renamed in your governed BigQuery datasets and, because that breaks downstream queries hard, pages the on-call via PagerDuty and posts…
PR-time Snowflake schema contract check on dbt model changes
When a pull request changes a dbt model, it compares the model's declared output columns against the live Snowflake table it will replace and blocks the merge with a GitHub check…
Agent-triaged warehouse drift with impact analysis and runbook update
On a webhook from your warehouse audit log, an agent investigates the changed column, traces which downstream models and dashboards depend on it.
Cross-warehouse replication schema mismatch reconciler
Compares the column shape of mirrored tables between BigQuery and Snowflake and, when a replicated table has drifted out of sync between the two, opens an Asana task for the data…
Run it inside a business
This workflow drops into a full company template. Import the org, and this is one of the playbooks its agents run.
Run this workflow in your colony.
14-day trial. No DevOps. No Sales call. Provisioned in under a minute.