Overview:

🎧 Talk: How to Build a Research Agent Using Pantic AI

👩‍🔬 Speakers: Angelina and Mehdi

📢 Hosted by: TwoSetAI YouTube Channel

📺 Watch on YouTube: Link to Video

⏱️ Full Talk Length: ~20 mins

⏱️ Summary Reading Time: ~4 mins

🧠 Key Concepts

1. Why Build a Research Agent?

• Research and summarisation are the most common AI agent use cases across industries (per LangChain survey).
• The goal: create a tailored, lightweight research assistant with minimal external dependencies.
• Fewer dependencies = more maintainable and secure software.

2. The Research Agent Workflow

• The agent begins with a user-defined topic.
• It generates a search query, runs a web search, gathers sources, and then summarises the results.
• It reflects on gaps and iterates up to a set number of loops (e.g. 2).
• Finally, it generates a consolidated summary with source links.

3. Building Blocks of the Agent

• Uses Pantic AI (an agentic framework) and LiteLLM to support multiple LLMs (GPT-4o, Claude, Gemini, etc.).