Wocie, the AI data-analyst
September 2025 - current
Creation of an agentic chatbot to make accessible both a relational database and a knowledge base.
This tool is able to answer any data-related question about the business, without the need for a human data analyst. Management can generate any realtime overview of operations by themselves. Privacy is guaranteed by only sharing the data-model with LLMs, never the data itself. Besides data related questions, it can also give answers based on a knowledge base.
My Contribution
- Architecture
- Implementation
External Links
Highlights
- Natural language questions are turned into an SQL query based on DDLs and lists with company terminology
- Easy data export
- GPT-like intuitive interface
- In close collaboration with the lead data analyst we succeeded in making Wocie answer almost all questions he would normally get
- Questions about company policy and the knowledge base are routed to a RAG Agent, where relevant FAQ cards are selected using cosine similarity
- The chatbot is trained to say 'I cannot answer this question with the information provided to me', preventing creative but unapproved answers
Users
- Almost all departments, from customer service representatives to management and directors
Technology Stack
algorithmsLangChainLLMFastAPIPythonRAGVector Database