This course is a complete, hands-on guide to building real-world AI agents and deep research systems using Google Gemini, LangChain v1, MCP, and modern RAG techniques. You will start from the absolute basics of AI agents and slowly move towards building advanced autonomous multi-agent systems used for deep financial research. The course is designed in a progressive way so that beginners can follow along, while experienced developers will still learn advanced production-grade patterns. The focus of this course is not only theory. You will build everything step by step using Python notebooks, real APIs, real documents, and real data pipelines.
What this course covers
You will first understand what an AI agent really is. You will learn different agent patterns, how agents reason, how they take actions, and how to choose the right agent design for a real project. You will then set up Google Gemini AI Studio and LangSmith properly. This includes creating API keys, understanding pricing, rate limits, and tracing agent executions so you can debug and monitor your agents like a professional.
After that, you will go through a complete Gemini and LangChain bootcamp. You will learn how to use Gemini models in Python, how messages work internally, how streaming responses work, how multimodal inputs are handled, and how to use tools, function calling, reasoning mode, grounding, and context caching to reduce cost and improve performance.
Once the foundations are clear, you will move into LangChain agents. You will build agents with memory, state management, summarization middleware, fallback models, PII protection, planners, streaming responses, and structured outputs using Pydantic. The course then introduces MCP through a finance use case. You will connect external MCP servers like Yahoo Finance, load them as LangChain tools, and build a complete stock research agent with structured prompts and planners.
Deep RAG and Multimodal Finance Systems:
- A large part of this course focuses on Deep RAG systems for finance.
- You will learn why multimodal RAG is hard, what problems occur with PDFs, tables, images, and long documents, and how to design a reliable deep RAG pipeline.
- You will extract data from financial PDFs using Docling. This includes converting PDFs to markdown, extracting tables with context, tracking page numbers, extracting images, and validating data integrity at scale.
- You will then generate accurate image descriptions using multimodal Gemini models and store those descriptions in markdown so everything can be handled in a single text-based pipeline.
- Next, you will ingest large amounts of multimodal data into Qdrant vector database. You will learn dense search, sparse search, hybrid search, metadata filtering, de-duplication using file hashes, and best practices for chunking and retrieval models.
- On top of that, you will build advanced retrieval pipelines using hybrid search and cross-encoder re-ranking for better answer quality.
