<- Back to portfolio

Case study

Dataverse - Conversational AI Analytics

Upload a dataset, ask in plain language, get SQL and charts - no analyst in the loop.

Type
Independent build, public repo
Stack
LLM, Python, Streamlit, Pandas, SQL Generation

Problem and constraints

Business users needed ad-hoc analysis but lacked SQL skills, so analysts spent hours writing queries for simple questions like 'show me sales by region last quarter.'

Approach

Built a multi-agent system with cleaning, enrichment, SQL generation, and visualization specialists. Users upload any dataset and ask questions in natural language; the system handles data cleaning, generates SQL, and returns charts.

Architecture

Diagram of the Dataverse request path: a dataset uploaded as CSV, Excel or Parquet, the cleaning and enrichment agents, the SQL generation agent, and the visualization output.
Request path from an uploaded dataset through the specialist agents to SQL and charts.

Measured result

Enabled non-technical users to explore data without analyst intervention, removing the repetitive reporting requests that made up the ad-hoc queue.

What Value Scope and source
commits shipped 191 single public repo, multi-agent analytics app
specialist agents 5 cleaning, enrichment, SQL generation and visualization over uploaded CSV, Excel and Parquet

Stack

LLM Python Streamlit Pandas SQL Generation