Designing Multi-Agent Deep Search Systems: A Practical Architecture Workshop for Planning, Tool Use, Validation, Merging

Designing Multi-Agent Deep Search Systems is a 1.5–2 hour technical workshop on how to architect deep search agents that go beyond simple retrieval.

In this session, we will break down how to design an agentic workflow that can plan a search, use tools, collect evidence from multiple sources, validate information quality, handle contradictions, reason about dates and freshness, merge findings into a structured output, and improve across iterations.

The workshop is architecture- and design-focused. We will focus on the core components, design decisions, tradeoffs, and patterns needed to build reliable deep search systems, without going deep into implementation code.

What we will cover

• What deep search agents are
  • LLMs vs agents vs deep search agents
  • Why deep search is different from normal search or basic RAG
• Multi-agent architecture
  • planner agents
  • executor agents
  • validation agents
  • merging agents
  • improvement/feedback agents
• Planning and iterative search
  • breaking a topic into sub-questions
  • deciding which sources and tools to use
  • knowing when to continue searching and when to stop
• Tool design for deep search
  • search tools
  • scraping tools
  • OCR and image understanding tools
  • video-to-text tools
  • social/data enrichment tools
  • think tools for deeper reasoning
• Tools vs MCPs
  • what MCPs are
  • when they are useful
  • when normal tools are enough
• Validation and source quality
  • source reliability scoring
  • official vs unofficial sources
  • handling conflicting claims
  • using source authority, freshness, and corroboration
• Temporal reasoning
  • managing publication dates, event dates, update dates, and retrieval dates
  • avoiding outdated or mixed-time conclusions
• Merging information from multiple sources
  • deduplication
  • conflict resolution
  • preserving source provenance
  • producing a clean final structured result
• Context and memory management
  • short-term working context
  • long-term memory
  • storing lessons from previous iterations
• Evaluation
  • how to evaluate deep search agents end to end
  • what to measure beyond retrieval quality
  • common failure modes and design tradeoffs

Deliverables

When you join the workshop, you will get:

• 1.5–2 hour live technical workshop
  A detailed walkthrough of how to design a multi-agent deep search system from an architecture and system-design perspective.
• Full workshop recording
  Access to the replay after the live session.
• Complete slide deck
  The full presentation used during the workshop, including architecture diagrams, workflows, and design patterns.
• Technical workshop notes / handbook
  A written reference guide summarizing the main concepts, architecture decisions, and practical design patterns covered in the session.
• Multi-agent deep search architecture blueprint
  A clear system blueprint showing the planner, executor, validation, merger, memory, and improvement layers.
• Tool design reference
  Examples of useful tools for deep search systems, including search tools, scraping tools, OCR/image tools, video-to-text tools, enrichment tools, and think tools.