Module 5 of 5
Agentic Workflows
Building autonomous research agents with tool use, ReAct patterns, and multi-step orchestration: the frontier of what LLMs can do for your research.
Agents & Tool Use
An LLM agent is a system where the model doesn't just generate text but takes actions: calling APIs, querying databases, running code, or browsing the web. This transforms the model from a text processor into an autonomous research assistant.
Definition
ReAct Pattern
A prompting framework where the model alternates between Reasoning (thinking through what to do next) and Acting (executing a tool call), using the results of each action to inform subsequent reasoning. This creates a transparent, step-by-step problem-solving loop.
We cover the building blocks of agentic systems: function calling conventions, the Model Context Protocol (MCP) for tool integration, and frameworks like LangGraph for orchestrating multi-step workflows.
Orchestration & Workflows
Real research tasks rarely fit into a single LLM call. Agentic workflows compose multiple steps: retrieval, analysis, validation, reporting: into pipelines that can process data with minimal human intervention.
We examine practical patterns: how to design reliable multi-step pipelines, handle errors and retries, implement human-in-the-loop checkpoints, and ensure reproducibility in workflows that involve stochastic LLM outputs.
Key Takeaway
Agentic workflows represent the frontier of LLM applications in research. They can automate complex analysis pipelines, but they also amplify errors: a mistake in an early step propagates through the entire chain. Careful design with validation checkpoints is essential.
Capstone Project
The final module brings everything together. You will design and implement an end-to-end research pipeline using LLMs, drawing on the concepts and techniques from all five days of the course.
Think about a research question from your own work. How could you decompose it into a multi-step pipeline using the techniques covered in this course? Which modules would each step draw on?
Capstone ideas: Automated content analysis of parliamentary debates, RAG-powered policy document analysis, simulated focus groups for survey pretesting, or an agentic literature review pipeline.