Project

RAG Guess Who

Chat-based character guessing game built with roleplay prompts, retrieval, and ingestion tooling

Demonstration of a simple RAG implementation as a Guess Who-style game. Built with Next.js and LangChain, the app uses a streaming conversational UI powered by the Vercel AI SDK so users can chat with a hidden character and infer the identity from retrieved context. The pipeline scrapes web pages, cleans HTML, creates a corpus of text documents, and chunks that content into JSON files as part of a simple corpus ingestion workflow before transforming it into vector embeddings for Pinecone. At runtime, the app performs vector search and semantic retrieval to ground responses. It should be acknowledged: famous characters are already well known by LLMs, so a stronger real-world use case would be non-famous people or internal organizational knowledge.

  • Next.js
  • React
  • TypeScript
  • assistant-ui
  • Vercel AI SDK
  • LangChain
  • Pinecone
  • RAG
  • LLM Prompt Design
  • Ingestion Pipelines