Digital Twin

A chatbot that answers questions about the career on this page, in the visitor’s language. Demonstrates an agentic RAG system: several agents filter each question, retrieve the documents, write a professional answer, and send push notifications, while semantic vector search and BM25 lexical search run concurrently and are reranked with Cohere for precision.

How I built it

Designed an orchestrated multi-agent workflow with LangGraph and OpenAI agents, breaking the conversation into specialised steps. Each agent can access the tools it needs, while structured outputs and guardrails provide consistency and control over the responses.

To improve retrieval quality, implemented a hybrid RAG approach combining BM25 lexical search and semantic vector search. Relevant documents are retrieved from the Chroma database and then reranked with Cohere before being provided as context to the agents.

Built the backend with FastAPI, exposing the agentic RAG pipeline through APIs that connect the AI layer with the frontend.