Data Lab on every pipeline

Interactive analysis fed by the same AI agents: profile, detect anomalies, visualize, then send the session to training. The lab is not a sidecar — it is the output of the pipeline.

Agents on the same output

Profiling

Column stats, types, and quality scores from sample rows before any LLM call.

Anomaly & quality

Drift, outliers, and scorecards the same crew that built the pipeline can explain.

Visualization

Charts and artifacts generated against the live profile, not a disconnected BI extract.

Ask in English

NL queries run as guided analyses on the session — then one click to ML Lab.

Why it stays first-class

Same pipeline output feeds Data Lab — no warehouse dump or notebook setup
Deterministic profiling first: nulls, distributions, quality scorecards
Agents for anomalies, correlations, and natural-language questions
Visualizations and artifacts stay attached to the session
Promote findings back into the pipeline or send the session to ML Lab
Reuse the latest session per pipeline so analysis stays cheap and current

Questions about this use case

No. Data Lab opens on the pipeline’s sample rows and profile. The latest session per user and pipeline is reused so you don’t bootstrap a new environment every time.

Analysis that can become a model

When the profile is good enough, send the session to ML Lab. Target scan, baseline, deploy — still on the same pipeline.