Build an AI Agent (OpenAI, LlamaIndex, Pinecone & Streamlit)

Create a tools module using the archive library to build a default arXiv API client and fetch papers via a title-based all query, returning title, authors, summary, and URLs.

Build a pdf downloader tool that fetches a pdf url with requests, creates a papers directory if needed, saves the file to a full output path, and handles errors.

Define the embed and LLM models in a constants file, load the OpenAI key from the environment, and configure embed three large and GPT-4 mini.

Build a knowledge base index from language model papers by turning papers into documents, embedding with a large model, chunking with overlap, and persisting the index in a local folder.

Build a rag query engine tool by loading the local index, configuring a Lama index query engine with five-vector similarity, and refining answers using text and refined templates from prompts.

Defines two tools for downloading pdfs and fetching archive papers, builds a reactive agent with two-stage reasoning to retrieve multimodal models papers, including titles, summaries, authors, and download links.

The agent downloads all cited papers by remembering completed tasks, then uses a download pdf tool to fetch multiple papers and even fetch new arXiv results.

Learn iterative prompt engineering to control an AI agent’s downloads, refining prompts to download papers one by one and only when explicitly requested by the user.

Create an index manager class to fetch papers, build documents, and assemble a reusable index for a streamlit app, then prepare the agent class for future use.

Build an agent class that wires an index and an LM model into a query engine with rag tool, pdf download tool, and fetch archive tool, plus a chat method.

Build a Streamlit chat UI to interact with an AI agent, using session state, cached initialization, and a simple user–assistant chat flow that fetches papers via an archive tool.

Extend the index manager to create a Pinecone-backed index, load the embed model, upsert vectors with a Pinecone vector store, and retrieve the index for paper embeddings.

Configure pinecone index in a streamlit app using archive research to enable the agent to retrieve papers on multimodal models, including zero resource speech translation and long form speech generation.

Deploy your app to the Streamlit community cloud by linking GitHub, creating an app, and configuring secrets with OpenAI and Pinecone keys; generate a requirements.txt from the pdm log.

Celebrate completing this course on building AI agents by applying your knowledge to real world projects, sharing achievements with the community, and staying curious about AI advancements.