Datavisual.studio
Upload any dataset, ask any question, and get a structured report where four leading AI models independently analyse your data alongside live internet research — then a chairman model synthesises a final, cited answer.
Overview
Datavisual.studio turns a raw dataset into an AI-audited research report. Upload a CSV, Excel, or JSON file — or connect a SQL database or REST API directly, Power BI-style — and either build a live dashboard from it in one click, or ask it a question in plain language.
For questions, the platform profiles the data, generates charts automatically, and searches the internet for current context. Four AI models then analyse everything independently, review each other's reasoning, and a chairman model produces a single synthesised answer with comparison tables, source citations, and a live activity panel showing the whole pipeline in real time.
What It Does
Flexible data in
CSV, Excel, JSON upload, or a direct connection to PostgreSQL, MySQL, SQLite, or a REST API.
One-click dashboards
Metric cards, auto-generated interactive charts, entity comparison, a sortable table, and CSV/HTML export — no AI run required.
Chat-driven editing
"Add a pie of revenue by product", "remove the histogram", "research this and pin the findings" — edits apply to the live dashboard in place.
Four-model council
Independent analysis from four AI models via OpenRouter, each reviewing the others' reasoning before a chairman model synthesises the final answer.
Full report
Comparison tables, source citations, and a live Activity Panel showing the research and reasoning pipeline as it runs.
Follow-up Q&A
The chairman model answers follow-up questions using the full session context, not a cold restart.
How Predictions Work — An Example
Upload elo_ratings_wc2026.csv (4,683 rows, 23 columns) and ask: *"Which team has the highest probability of winning the 2026 FIFA World Cup?"*
pandas detects the ELO rating column, the country entity column, and a snapshot-date time column, then computes a form index weighted toward recent periods. One model runs 10,000 bracket simulations, another runs 10,000 Poisson simulations with Dixon-Coles adjustment, and the two are ensembled. In parallel, live research pulls current odds and expert forecasts. All four council models see the dataset, the predictions, and the research together, peer-review each other's reasoning, and the final number is a weighted blend of the simulation ensemble, the internet research, and the council consensus — explained, but not altered, by the chairman.
Stack
App Router frontend, SSE stream handling, interactive charts via react-plotly.js
Upload/connect endpoints, the analysis pipeline, and SSE streaming to the frontend
Data profiling, auto chart generation, and the prediction engine (ELO Monte Carlo, Poisson/Dixon-Coles)
Single-key access to the four-model council, peer review, and chairman synthesis
PDF report export, with automatic HTML fallback when unavailable