An Introduction to Creating Agentic Applications
Details
Background
The tooling around LLMs is evolving super fast and at times it can be hard to keep up. LLMs unlock an entire new world of tasks for computer programs that previously could only be completed by trained humans. Now, we can wire together systems of agents to complete the complex tasks. Throughout the talk, I will alternate between introducing new concepts and demonstrating how we can add them to a small example application.
Summary
I will cover the following concepts:
- The OpenAI API: acquiring a token and making your first API request to OpenAI via curl.
- Langchain: an importable library for Python and other languages that simplifies the process of creating agents, regardless of the underlying provider.
- Structured output: Forcing LLM outputs to take the shape of the data that our program expects.
- Tool Calls: a way to give agents the ability to take actions either fully autonomously, or with human approval, depending on the needs of the application. This usually means calling functions within our program but can also mean spawning other agents.
- RAG (Retrieval Augmented Generation): a technique for giving agents access to large data stores without polluting their context window.
Benefits
You will learn about a fast-moving area that is changing how we think about computer programs and pushing the limits of what they can do in an approachable, practical way.
Bio
Adithya Laakso is an aspiring startup founder developing an agentic application in the education space.