Challanges

While Large Language Models (LLMs) are not naturally suited for statistical analysis or plotting, the main challenge was integrating these capabilities into the LLM workflow. Leveraging Python’s asyncio framework was highly advantageous for the application’s performance, though some libraries’ lack of asyncio support posed a significant hurdle.

Solutions

The solution is an asyncio-driven application leveraging LangChain and its SQL Toolkit to connect to a database, which is automatically populated by monitoring a directory for new measurement files. The measurement file and data format are fixed yet designed for easy adaptation to other structures. The AI’s functionality was enhanced with a custom layer that identifies when plotting, FFT, or advanced statistical analysis is required, generates the corresponding content locally without sending much tokens to the AI service, integrates it into the chat interface, and appends it to the agent’s history context. This enables users to seamlessly discuss the results with the agent through text as well.