Building LLM-Powered Applications

• Understand the capabilities, limitations, and tradeoffs of large language models and select appropriate models and APIs for different applications.
• Integrate LLMs into web applications using a secure backend architecture.
• Apply prompt engineering techniques, including system prompts, few-shot prompting, chain-of-thought, and structured outputs.
• Build conversational AI features with streaming responses and conversation history.
• Implement Retrieval-Augmented Generation (RAG) to connect LLMs with your own data.
• Use embeddings, vector databases, chunking, and retrieval to build grounded AI applications.
• Integrate multimodal LLMs to process images and documents.
• Implement structured extraction and function/tool calling.
• Compare RAG and fine-tuning and understand when each approach is appropriate.
• Evaluate LLM responses using practical techniques such as LLM-as-judge.
• Apply responsible AI practices to address hallucinations, prompt injection, and unsafe outputs.
• Design and build an end-to-end LLM-powered application that incorporates retrieval, evaluation, and a safety layer.

Topics Covered

• Large Language Models and Generative AI: Explore LLM capabilities, APIs, model selection, and the fundamentals of integrating LLMs into modern applications.
• LLM Application Development: Learn the development stack and architecture for building web applications powered by large language models.
• Prompt Engineering: Apply system prompts, few-shot prompting, structured outputs, and other techniques to improve LLM performance.
• Conversational AI: Build LLM-powered chat applications with streaming responses and conversation history.
• Retrieval-Augmented Generation (RAG): Use embeddings, vector databases, chunking, and retrieval to ground LLMs in your own data.
• Multimodal LLMs: Build applications that use LLMs to process and understand images and documents.
• Advanced LLM Application Patterns: Explore structured extraction and function/tool calling to extend the capabilities of LLM-powered applications.
• RAG and Fine-Tuning: Compare RAG and fine-tuning approaches and understand when each is appropriate for an AI application.
• LLM Evaluation: Learn practical approaches for evaluating LLM output quality, including LLM-as-judge techniques.
• Responsible AI and LLM Deployment: Address hallucinations, prompt injection, output filtering, and other considerations for deploying LLM applications responsibly.

Course Details and Next Steps

• Course typically offered: Online in Summer and Winter quarters
• Prerequisite: Students should have a foundational understanding of AI and machine learning concepts and be comfortable writing code. Experience with JavaScript, TypeScript, or web development is helpful.
• Next Step: Upon completion of this course, consider taking courses in the Technical Aspects of Artificial Intelligence certificate program.
• Contact: For more information about this course, please email unex-techdata@ucsd.edu.