Build Your Own Deep Research Agent with Ivan Leo (ex-Manus)
OpenAI, Anthropic, Perplexity, and Google all have deep research products. You give them a question, they fan out across dozens of sources, synthesize what they find, and return a comprehensive report with citations.
They're also black boxes.
In this session, you'll build one from scratch. And we'll stress-test it on one of the messiest, most complex document corpuses available.... to be announced ;)
Join Ivan Leo (ex-Manus) and Hugo Bowne-Anderson (Vanishing Gradients( for a live, hands-on build.
Starting from a blank Python file, we'll construct a multi-agent research system that can:
Break complex questions into parallel research threads
Search, retrieve, and read documents autonomously
Manage context across hundreds of sources without drowning
Synthesize findings into coherent, cited reports
What we'll cover
Multi-agent orchestration: Build a supervisor that delegates to specialized sub-agents with isolated context windows, the same pattern Anthropic uses in Claude's deep research
Context engineering: Implement compaction, structured note-taking, and just-in-time retrieval. Learn why "just dump it in the context window" fails at scale
Document ingestion: Connect your agent to real document corpuses: PDFs, web pages, structured and unstructured data
Search strategies: When to go broad vs. deep. How to let the agent decide. Parallel tool calling for 10x speed
Synthesis and citation: Turn raw research into coherent reports with proper attribution
Deep Research Evals: How to evaluate your deep research agent
Who this is for
Engineers who want to understand how deep research actually works under the hood
Builders who need to search and synthesize across large document sets
No prior agent experience required.
Register to join live or get the recording afterwards.