SpringBoot kitchen sink for ChatGPT, vector RAG, and MidJourney

What it does FunAi is a SpringBoot backend that acts as a thick integration layer around OpenAI’s APIs. It offers multi-session ChatGPT and GPT-4 chat with context caching, a retrieval-augmented document Q&A pipeline over PDFs and Office files via Pinecone or Milvus, and hooks for MidJourney and Stable Diffusion image generation. The codebase also bundles AI customer service, prompt management, user tiering, and text adventure game logic.

The interesting bit Despite billing itself as a learning project, it tackles the unglamorous plumbing that production wrappers need: a polling scheduler to rotate OpenAI API keys and dodge free-tier rate limits, recursive text chunking for embedding-based search, and Guava Cache to keep conversation context responsive. It is essentially a maximalist reference architecture for Java developers wondering how to wire LLMs into a traditional web stack.

Key highlights

• Wraps OpenAI chat, embeddings, and image generation behind a SpringBoot backend with SSE streaming and multi-session history.
• Document Q&A pipeline: parses PDFs and Office files, recursively chunks text, vectorizes via OpenAI embeddings, and retrieves semantically similar passages from Pinecone or Milvus.
• API key pool with a polling scheduler to distribute requests and detect invalid or exhausted keys.
• Includes gamified text adventures, AI customer service, and dynamic prompt libraries with role-based access.
• Runs only with overseas network access or an overseas server due to OpenAI API reachability.

Caveats

• Several headline features—human handoff for customer service, resume analysis, spoken practice, and the “devil community”—are still marked TODO in the README.
• The matching frontend repository is not yet open-sourced, so this release is backend-only.
• The author explicitly restricts commercial use without permission, which limits its utility as a drop-in foundation despite the Apache-2.0 license.

Verdict Java developers who want a sprawling, real-world reference for gluing OpenAI services into a monolithic web app will find plenty of wiring to study. If you need a lean, library-style SDK or a fully open, production-ready product, look elsewhere.