Projects
02 — Evidence-based multi-agent research
DeepScout
DeepScout is an open-source system for structured research grounded in sources and evidence. It breaks an objective into tasks, coordinates agents, uses hybrid retrieval, and measures coverage to investigate material gaps. It includes Human-in-the-Loop controls, persisted evaluations, and versioned adaptive policies with monitoring and rollback.
Real interface
01 / 04
Research flow connecting orchestration, planning, specialised agents, verification, synthesis and report generation.
- It turns an objective into requirements and DAG-linked tasks, then coordinates research agents within explicit budgets and operational limits.
- It maintains a verifiable chain from requirement to claim, quotation, snapshot and source; coverage detects gaps and can trigger targeted corrective research.
- It combines Human-in-the-Loop controls, persisted evaluations and cited reports without treating unproven model text as fact.
- Adaptive policies are versioned and bounded: experiments, monitoring and rollback improve runtime behaviour without training models or autonomously changing source code.
Stack
PythonTypeScriptNext.jsFastAPIPostgreSQL + pgvectorLangChain / LangGraphDocker