A dataset analyst you can trust.
DataVerse AI was built around a simple idea: an AI data analyst is only useful if its numbers are correct. So every figure is computed deterministically, and the language model is kept on a short leash — it explains results, it never generates them.
Deterministic-first
Any number a user sees is computed with Pandas or scikit-learn. The LLM is optional and only polishes narration — it can never fabricate a metric.
Exactly two agents
A DatasetAgent owns ingestion and integrity; an AnalystAgent owns analysis and explanation. No sprawling swarm, no hidden state.
Concise by design
Reports stay 1–2 pages: at most two charts, three recommendations, and a hard rule that no fact is ever repeated twice.
Runs anywhere
No database, Supabase, or API keys required. With nothing configured it falls back to local storage and offline Mock mode.
How a request flows
- 1
Upload & validate
Secure file handling, CSV/XLSX parsing, header normalisation.
- 2
Profile & quality
Type inference, semantic mapping, missing/duplicate/outlier scan.
- 3
Metrics & EDA
Business metrics, trends, correlations — all deterministic.
- 4
Predict & explain
Optional model + SHAP, gated on enough rows; skipped with a reason otherwise.
- 5
Compose & render
De-duplicated sections assembled into a compact HTML/PDF report.
Built with
Backend
FastAPI · Pandas · scikit-learn · SHAP · ReportLab
Frontend
Next.js 15 · React 19 · Tailwind v4
Persistence
Local filesystem by default · optional Supabase
LLM
Optional narration polish only · offline Mock mode