Hanse: Enterprise LLMs & Data Querying
Week 3 of the High Digital AI Initiatives. Today, we look at how Generative AI is replacing complex SQL queries in the trade data and supply chain analytics sector.
Quick Answer: Can you query a database using an LLM?
Yes. High Digital’s Hanse application allows users to query trillions of trade data records using a natural language chat interface. Powered by a locally hosted, specially trained Meta Llama LLM, it translates user intent into rapid database queries without requiring SQL knowledge.
Why Deploy a Local LLM for Enterprise Data?
When dealing with large amounts of trade data, or customer data, routing queries through public APIs (like OpenAI) introduces unacceptable latency and severe data privacy risks.
To solve this, Hanse utilises a local deployment of Meta Llama. We trained this model specifically on our proprietary data schemas and trade data use cases. This architecture ensures the AI behaves strictly according to predefined constraints, maintains zero-trust data security, and delivers near-instantaneous response times by eliminating external network calls.
The Tri-Fold Output Engine
When a user asks a complex question in natural language (e.g., “Compare the volume trends for X sector vs Y sector over the last 3 years”), the LLM processes the query and Hanse’s rendering engine returns the output in three interactive formats:
- Data Stories: The LLM generates a contextual narrative explaining the anomalies and trends found within the dataset.
- Dynamic Charts: The system auto-generates visual representations, including column, line, and pie charts using the retrieved data arrays.
- Exportable Matrices: The raw queried data is formatted into clean, highly structured tables ready for immediate CSV/Excel export.
Next Week in the AI Series: We step into the field to show how edge-based computer vision is eliminating manual data entry.
We love tackling complex enterprise data solutions and AI projects. If you have one drop us a line
