CS50 AI Workshop

This workshop is designed to introduce you to the capabilities of OpenAI's APIs, including the Responses, Embedding, and File Search APIs, with hands-on demonstrations and code examples.

Slides for this workshop are available here.

Requirements

• Python 3.x
• OpenAI Python Library (installation guide below)
• OpenAI API Key
• Internet Connection

Installation

Before we dive into the demos, please ensure your environment is set up with the necessary software and libraries:

# Install the OpenAI library and other dependencies
pip3 install -r requirements.txt

Demo 1: Responses API

This demo illustrates how to utilize the Responses API to create an interactive chatbot.

Key Features

• Developer Message: Sets the context for the AI (e.g., "You are a friendly and supportive teaching assistant for CS50. You are also a cat.")
• User Interaction: Accepts user input to simulate a conversation.
• API Integration: Utilizes the responses.create method to generate responses based on the conversation history.
• Streaming Responses: Demonstrates how to handle streaming responses with ResponseTextDeltaEvent.

Demo 2: Text Embeddings and Semantic Search

This demo showcases the use of OpenAI's text embeddings to perform semantic search, enabling the identification of the most relevant information chunk in response to a user query. This technique can significantly enhance the way educational content is queried and retrieved, making it a powerful tool for educators and students alike.

Key Features of Demo 2

• Text Embeddings: Illustrates how to generate and utilize text embeddings using OpenAI's embeddings.create method.
• Semantic Search: Demonstrates how to compute similarity scores between embeddings to find the most relevant content.
• Integration with Responses API: Combines the result of semantic search with the Responses API to generate contextually relevant responses.

Usage Notes

• Pre-computed Embeddings: Before running this demo, ensure you have an embeddings.jsonl file containing pre-computed embeddings for various content chunks relevant to your subject matter.
• Custom Model Selection: You can experiment with different models for embeddings to suit your content and accuracy requirements.

Demo 3: Responses API with File Search

This demo showcases how to use the Responses API with the file_search tool and a vector store to provide tailored responses based on specific data files. It is particularly useful for building retrieval-augmented generation (RAG) applications for events, courses, or research projects.

Key Features

• File Search Tool: Guides you through uploading files, creating a vector store, and using the file_search tool to answer CS50 or computer science-related questions.
• Data File Utilization: Demonstrates how to upload and associate data files with a vector store to enrich the model's responses.
• Streaming: Shows how to stream responses incrementally using ResponseTextDeltaEvent.

Usage Notes

• Data Preparation: Before running the demo, ensure your FILES_DIR points to the directory containing relevant files you wish to search over. We have pre-configured the use of lecture transcripts in the example.
• Customization: You can customize the instructions, behavior, and tools to fit various educational or research contexts.

Demo 4: RAG with ChromaDB

This demo showcases how to build a local retrieval-augmented generation (RAG) pipeline using ChromaDB as a persistent vector store. Instead of uploading files to OpenAI, it processes SRT subtitle files into time-based chunks, embeds them locally with OpenAI's embedding model, and stores them in a ChromaDB collection for fast semantic search.

Key Features

• SRT Chunking: Parses subtitle files and merges consecutive captions into ~30-second chunks, preserving start and end timestamps.
• Local Vector Store: Uses ChromaDB with persistent storage so embeddings are computed once and reused across sessions.
• OpenAI Embeddings: Embeds chunks using the text-embedding-3-large model via ChromaDB's OpenAIEmbeddingFunction.
• Streaming: Streams responses incrementally using ResponseTextDeltaEvent.
• Source References: Each response includes clickable YouTube links that jump to the exact lecture timestamp.

Usage Notes

• Ingestion: Before chatting, run ingest.py to process all SRT files and populate the ChromaDB collection. This only needs to be done once (or re-run if the data changes).
• Chat: Run chat.py to start the interactive Q&A. Use the -v or --verbose flag to display the retrieved chunks.
• Customization: You can adjust chunking parameters (CHUNK_MIN_DURATION, CHUNK_MAX_DURATION), the embedding model, and the number of retrieved chunks in config.py.

# Step 1: Ingest SRT files into ChromaDB (one-time)
cd examples/rag/chromadb
python3 ingest.py

# Step 2: Start the interactive chat
python3 chat.py

# Or with verbose mode to see retrieved chunks
python3 chat.py -v