Skip to content

MAS Agents Research 1.0

Cyberpunk-themed dual-mode research orchestration system using specialized AI models for deep-dive analysis and map-reduce synthesis.

MAS Agents Research 1.0 workflow: Research question → Planner agent → Research agents → AI validation → Research report. Multi-agent orchestration · ChromaDB memory.

Deep Research Swarm · Role: Architect · Team: Solo

Architecture

  • Dual-mode: Agentic Deep Dive + Map-Reduce Military Swarm
  • 6 specialized models (Planner, Researcher, Reasoner, Drafter, Validator)
  • Self-healing Engineer Agent generates Python tools at runtime

Engineering approach

  • Runtime AST validation & security scanning
  • Semantic tool memory via ChromaDB
  • Vector-ID dispatch pattern (57% cost savings)

Scale and implementation

  • Processes 50-100+ web sources per query
  • 73% speed improvement via parallel dispatch

Technology stack

  • LangGraph
  • ChromaDB
  • Python
  • DeepSeek
  • Llama 3.3

Related projects

Build with Fern