Hands-On Workshop: Building a RAG-Powered Q&A Bot

About This Course

Aim

Workshop Structure

• Environment & Dependencies Setup

  Ensure the environment is properly configured for the project by following these steps:

  • Spin up a Python virtual environment (venv) for isolation.

  • Install necessary dependencies using pip:

    bash

    pip install langchain faiss-cpu openai dotenv fastapi uvicorn


  • Create a .env file to store your API key for OpenAI integration.

  ---

  Document Ingestion & Embedding

  Integrate document ingestion and text embedding with the following process:

  • Write a script to load sample text/PDFs from the ./docs/ directory.

  • Chunk the texts into manageable segments (e.g., 500 tokens per chunk).

  • Use OpenAIEmbeddings to generate embeddings and store them in FAISS for efficient retrieval.

  ---

  Vector Store & Retrieval Function

  Set up the vector store and implement a retrieval function:

  • Initialize the FAISS index in your code for storing and searching vectors.

  • Implement the retrieve(query): function using index.similarity_search.

  • Perform quick tests by printing out retrieved chunks for sample queries.

  ---

  QA Chain Implementation

  Define a QA chain that utilizes the retrieved context for answering questions:

  • Create a prompt template that dynamically injects the retrieved context into the model.

  • Integrate the LLMChain (or RetrievalQA from LangChain) to process the query.
    Example code:

    docs = retrieve(user_question)
answer = llm_chain.predict(context=docs, question=user_question)


  • Test the system by asking 2–3 different questions to check if it retrieves accurate answers.

  ---

  API Endpoint / Minimal Interface

  Expose the solution via an API:

  • Scaffold a FastAPI app with a /qa POST endpoint.

  • Integrate the retrieve and llm_chain inside this endpoint for real-time querying.

  • Test the API via curl or Postman to ensure functionality.

  ---

  Testing, Debugging & Extensions

  Ensure the system is robust and extendable:

  • Handle cases where no results are found: return a message like “No context found.”

  • Experiment with different chunk sizes and k-values in retrieval for performance tuning.

  • Conduct a performance check: Measure latency for sample queries and optimize as necessary.

Who Should Enrol?

Important Dates

Workshop Outcomes

Meet Your Mentor(s)

Dr Shiv Kumar Verma

Professor

Sharda Institute of Engineering & Technology

more

Fee Structure

What You’ll Gain