Talk to sales
Workbook

Business Intelligence Developer

You own: the dashboards executives + product teams see. GeneFlow gives you new sources: ML cost, model health, drift incidents, prompt versions.

What's new for you

WasNow in GeneFlow
ML cost was in finance spreadsheetsgf_runs.cost_usd + gf_inference_logs.cost_usd queryable daily
Model health was an Engineering chartSame data, integrated into your BI tool
GeneAI features had no business metricsPrompt versions + eval scores → KPIs

Dashboards you can ship now

1. ML cost dashboard

Tiles (target: exec + ML lead):

  • Total monthly cost (training + inference + idle endpoints)
  • Top 10 models by 7-day cost
  • Trend of cost-per-prediction by model
  • Idle endpoint cost (endpoints with < 10 reqs/day)

Source views: fct_geneflow_cost_daily, dim_endpoints (see AE workbook).

2. Model production health

Tiles:

  • Active endpoints by status pie
  • Average p95 latency, last 24h, top 20 endpoints
  • 5xx rate trend
  • Open drift alerts by severity
  • Recent stage transitions (audit-derived)

3. GeneAI usage

Tiles:

  • Prompts in Production by team
  • Tokens / cost burned per prompt per day
  • Eval-set pass rate by prompt version (gf_eval_runs)
  • Active collab sessions right now

4. Lineage explorer (read-only)

Tile:

  • Free-form input: feature_group_id or dataset_id
  • Output: list of downstream runs + models + endpoints

Use the API directly:

GET /api/v2.1/geneflow/lineage/downstream?kind=feature_group&id=user_features_v2

Common gotchas

  • Replicating gf_inference_logs is expensive — high cardinality. Aggregate to daily / hourly rollups for BI, leave raw logs to AE / SQL ad-hoc.
  • Use the GeneFlow REST surface for live data — for dashboards that need < 5 min freshness, hit /api/v2.1/geneflow/endpoints/:id/metrics instead of warehouse.
  • Tenant filter — every dashboard parameter should include tenant_id.

Where to go next