Skip to main content

Project Exam

Assignment Details​

Assignment 1: RAG Agent System for FPT Policy

Description​

Build an automated question-answering system about FPT's internal regulations and policies. This system will serve as an internal tool to help employees quickly find information regarding company policies.

Objectives​

  • Implement a ReAct (Reason + Act) Agent using LangChain.
  • Build a Retrieval-Augmented Generation (RAG) pipeline.
  • Integrate a Vector Database for semantic search and retrieval.
  • Create a user-friendly interface for querying FPT policies.

Problem Description​

Develop a "RAG Agent System for FPT Policy" that takes user questions as input, searches through a provided set of FPT policy documents (converted to embeddings in a VectorDB), and uses a ReAct Agent to synthesize and provide an accurate answer based on the retrieved context.

Assumptions​

  • The necessary FPT policy documents and internal guides are provided in text format.
  • You have access to an embedding model and a standard Vector Database (e.g., Chroma, FAISS).
  • You have a valid API key for the LLM (e.g., OpenAI, Anthropic).
  • The environment is set up with Python and necessary libraries.

Technical Requirements​

  • Must use Python (version 3.8 or higher)
  • Must use LangChain library
  • Must use a Vector Database for document storage and retrieval
  • Must implement the ReAct Agent pattern (Reason -> Action -> Observation)
  • Must ensure the agent uses a "retriever" tool to access the VectorDB

Questions to Answer​

  1. Explain the "Operating Mechanism" of the ReAct Agent implemented in this project (Reasoning -> Action -> Observation).
  2. How does the Retrieval System ensure it fetches the most relevant text segments?
  3. Describe the complete processing flow (pipeline) from raw text data to the final user answer.

Estimated Time​

Estimated Time to complete: 180 mins