Generative-AI-Nanodegree

This repo contains a collection of project deliverables from my Generative AI Nanodegree course work.

Educational Objectives

A graduate of this program will be able to:

• Situate generative AI within the broader history, context, and applications of artificial intelligence and deep learning
• Adapt generative foundation models to perform tasks in novel contexts
• Use LLMs and prompt engineering to create a custom chatbot
• Use image generation models such as Stable Diffusion to perform image inpainting
• Build applications that use LLMs, implement semantic search with vector databases, and apply retrieval augmented generation techniques

Projects

1. Apply Lightweight Fine-Tuning to a Foundation Model

   1. Load a foundation model
   2. Identify and load a Hugging Face dataset for your particular task
   3. Utilize a state-of-the-art technique to adjust the foundation model's weights to meet the needs of your task, using a lightweight (AKA parameter-efficient) fine-tuning technique that improves performance faster and more efficiently

2. Retrieval Augmented Generation(RAG)

   1. Find and prepare a dataset that augments a foundation model's knowledge, from a source such as APIs, web scraping, or documents on hand
   2. Create a semantic search pipeline by implementing a custom Python vector similarity search algorithm to match user questions to relevant parts of the custom dataset
   3. Compose a custom query by combining the semantic search results with the user's question and send it to the foundation model

3. AI Photo Editing with Inpainting

   1. Create a segmentation mask by differentiating between the subject and background of an image and create a matrix of pixels indicating the locations of these two components
   2. Given a text prompt and the pixel locations of the subject or background, replace part of the image with an AIgenerated image
   3. Connect your inpainting pipeline to a web interface that allows users to upload their own images and specify their own text prompts udacity

4. Personalized Real Estate Agent

   1. Generate synthetic data using LLMs
   2. Embed property listing data in a vector database
   3. Perform semantic search over property listings against user preferences
   4. Design prompts and use RAG techniques to deliver personalized recommendations

Dependencies

requirements.txt

License

License

Credits

• udacity.com
• CODEOWNERS