Capstone: The Autonomous Research Assistant

Congratulations on reaching the Capstone! You have learned the theory, the infrastructure, and the ethics. Now, it is time to build.

Your task is to build a Multi-Agent Autonomous Research Assistant. This is not a simple chatbot. It is a system that can take a vague research topic, find the relevant information, and move through a reasoning loop until it produces a high-quality report.

1. The Project Requirements

To pass this course, your application must include:

A. RAG Capability (Module 5)

The system must be able to ingest a set of PDF documents (Technical Whitepapers).
It must store these in a Vector Database (Chroma or Pinecone).
It must perform Semantic Retrieval to find facts.

B. Agentic Reasoning (Module 7)

The system must use the ReAct Pattern.
If it cannot find the answer in the PDFs, it must have a Search Tool to look on the web.
It must have a Self-Correction loop to verify its own facts.

C. Evaluation & Observability (Module 9)

You must use LangSmith or similar to trace the agent's research path.
You must include a "Final Judge" step to grade the research report.

2. Choosing Your Domain

You can build this assistant for any of the following fields:

Medical Research: Analyzing new papers on a specific disease.
Legal Discovery: Identifying precedents in a set of court case files.
Product Comparison: Researching the top 5 competitors for a new software feature.

3. The Tech Stack (Recommended)

Language: Python 3.10+
Framework: LangGraph (for the complex reasoning loops).
LLM: Claude 3.5 Sonnet (for reasoning) and GPT-4o-mini (for summarization).
Vector DB: ChromaDB (Local-first).

4. The Final Output

Your Assistant should produce a Markdown report with the following structure:

Executive Summary.
Key Findings (Each must have a citation to a PDF or a URL).
Methodology (A brief description of how the agent searched).
Confidence Score (How sure is the agent of its findings?).

Summary of the Challenge

Over the next three lessons, we will guide you through:

Lesson 2: Designing the Graph (Architecture).
Lesson 3: Writing the Code (Implementation).
Lesson 4: Deploying and Evaluating (Production).

This is your moment to prove you are an LLM Engineer. Don't just follow the tutorials—innovate. If you find a better way to chunk the data or a cleverer way to prompt the judge, use it!

Exercise: The Project Vision

Which domain have you chosen for your Capstone?
List the 3 "Tools" your agent will need to be successful in this domain. (e.g., Google Search, PDF Reader, Calculator).

Tip: Choose a domain you are passionate about. The best AI systems are built by engineers who understand the data they are working with!