langchain-examples

This repository contains a set of Jupyter notebooks exploring different approaches to question generation and question answering using large language models (LLMs), retrieval-augmented generation (RAG), and tool-using agents. These are simple (yet fun!) examples with the aim to provide simple code examples.

📁 Notebooks Overview

1. question_generation_with_gemini.ipynb

Goal: Automatically generate multiple-choice questions from textual input.

• Uses Gemini model for generating questions.
• Generates questions based on Pokémon descriptions.

2. question_answering_with_RAG.ipynb

Goal: Evaluate question answering performance using RAG vs. a base LLM.

• Uses Qwen2.5-7B model via vLLM (locally hosted).

• Compares performance in two scenarios:

  1. Vanilla LLM (no retrieval)
  2. RAG-enabled (retrieval-augmented generation)

• Demonstrates how adding retrieval improves question answering accuracy

3. question_answering_agent.ipynb

Goal: Create and compare different agent strategies for question answering.

• Also uses Qwen2.5-7B via vLLM.

• Evaluates three different setups:

  1. Vanilla LLM
  2. DuckDuckGo Search Agent (uses internet search to retrieve answers)
  3. Custom Tool Agent (queries the Pokémon API as a tool)

• Showcases how external tools and APIs can extend LLM capabilities.

Requirements:

• Gemini API needs API key
• Qwen model, DuckduckGo and Pokémon API are all public and doesn't require any key or registration