Agentic analytics platform
Ask the business a question in plain English. Get the SQL, the numbers and the chart.
Lead AI Engineer
Questions about sales, inventory and logistics went into a reporting queue and came back days later. Business users needed to analyse data themselves, without writing SQL or waiting for an analyst.
How it works.
Intent understanding
The platform first works out what the user is asking (a trend, a comparison, a product question, a bottleneck) before touching any data.
LLM orchestration and tool-calling
An orchestration layer lets the model call the right tools in sequence: query generation, enterprise data retrieval and charting.
Validated Text-to-SQL
Generated SQL is validated before it runs, reaching about 88% execution accuracy on real business questions.
Auto-generated charts
Results come back as dynamically generated Python visualisations, so answers arrive as a chart, not a raw table.
More than 300 users analyse sales, inventory and logistics data in plain language. Ad-hoc report requests fell by about 50%, and turnaround went from days to minutes.
My role
- Built the platform as Lead AI Engineer
- Designed the LLM orchestration and tool-calling framework
- Owned Text-to-SQL generation and validation
Stack
- LLM tool-calling
- Text-to-SQL
- Python
- Data visualisation