ICCN Research Agent

ICCN Research Agent handles questions that need more than a short retrieved answer. It supports research into creative-economy actors, cities, provinces, policies, and comparisons between regional evidence and information held by ICCN.

The application has basic and deep research modes. Both begin with internal ICCN retrieval, then gather external evidence and synthesize a report. Basic mode uses a bounded plan, verification, and retry loop. Deep mode uses a longer ReAct-style reasoner and executor workflow for broader analytical tasks.

I developed the Research Agent end to end, including the research graphs, state model, internal retrieval, web-search integration, persistence, streaming API, and report-oriented outputs. AI Docs is a separate team project and is not part of this case study.

I designed and implemented the Research Agent as a separate analytical capability within ICCN AI. My work covered orchestration, research state, internal and web retrieval, evidence processing, report synthesis, persistence, API streaming, and operational tracing.

Basic mode plans sub-questions, searches internal ICCN knowledge, performs web search, verifies the evidence, and can retry up to three times before synthesis. Deep mode also begins with internal retrieval, then uses a ReAct-style reasoner and executor loop with a maximum of ten steps. Both modes generate related follow-up questions and persist their research artifacts.