Provisioning of an AI-Based Project Expert
Project Context: A legacy system used by the client featured outdated documentation and poorly documented code. The goal was to re-document the system and verify the accuracy of existing documentation. To achieve this, I developed a prototype for an AI-based „Project Expert“ that aggregates code and documentation into a vector database (ChromaDB) and generates focused responses to internal user queries via an LLM. This Agentic RAG system is utilized to assess document relevance and locate specific requirements within the implemented code. A critical requirement was the exclusive use of internal AI models to ensure full data compliance.
Technical Details:
- Ingestion of Code and Documentation:
- Development of a highly flexible process for converting project documents to Markdown.
- Technologies:
langchain4j,JavaParser, Apache POI, Flexmark, and Tika.
- Vector Embedding:
- Embedding of chunks into a ChromaDB vector database.
- Technologies:
langchain4j, Ollama withmxbai-embed-large.
- Prompt Provisioning & Interface:
- Deployment of a WebComponent frontend for website integration.
- McP-Server integration for chat applications.
- Technologies:
ChromaDB, Ollama.
My Contribution: Responsible for the feasibility study and conceptual design. Lead architect for the system design and implementation of the prototype. Provisioning of the application as a Docker container on the company’s internal development server.
Tech Stack: Java, Spring Boot, spring-ai, langchain4j, ChromaDB, Ollama, mxbai-embed-large, Docker, Apache POI, Tika, Markdown.