Descriptions:
This full workshop from the AI Engineer conference walks developers through building a deep research agent system from scratch, led by Louis-François Bouchard (CTO and co-founder of Towards AI, seven-year AI educator and researcher), Paul Iusztin (author of the LLM Engineer’s Handbook), and Samridhi, a machine learning engineer and technical writer.
The session opens with a critique of AI-generated content—particularly the generic patterns common in LinkedIn posts (vague generalizations, hallucinated model references, outdated information)—before presenting a multi-agent pipeline designed to replace shallow output with genuinely researched technical writing. The system centers on two tools: a deep research tool that queries multiple web sources and returns structured results, and a YouTube video analysis tool that passes video URLs directly to the Gemini API as file URIs, allowing the model to process video multimodally frame-by-frame rather than relying on pre-existing transcripts. This approach takes two to three minutes per video but captures visual information unavailable in captions.
These tools feed into a writing agent that produces LinkedIn-style posts with code snippets and accurate citations. The complete codebase is publicly available on GitHub. Instructors also share hard-won lessons from iterating the system with real students, covering prompt design, structured output formatting with Pydantic, and multi-agent orchestration patterns using the Gemini API.
📺 Source: AI Engineer · Published April 20, 2026
🏷️ Format: Hands On Build
