Gemini RAG Demo

Answers questions about Greek Government Gazette (FEK) PDFs with the document and page reference for every answer, so a reader can check the source instead of trusting the model. Retrieval uses Gemini embeddings, FAISS and LangChain over dense Greek legal text.

The problem Link to heading

Greek legal documents (ΦΕΚ) are dense and difficult to navigate. Finding specific information requires manual reading of multiple documents.

What I built Link to heading

Built a RAG pipeline that indexes PDF documents, stores semantic embeddings in FAISS, and uses Gemini 2.5 Flash to answer questions with source citations.

Implementation Link to heading

  • Implemented PDF ingestion, chunking and vector retrieval with Python, LangChain and FAISS.
  • Returned source document and page references alongside generated answers.
  • Included a Colab notebook for reproducing the workflow; citation correctness and abstention remain evaluation priorities.