shythu49/competeinsight

CompeteInsight is a research automation tool that takes a competitive analysis question and turns it into a full written report. You describe a product, name the competitors you want to study, set a research objective and the dimensions you care about (for example, pricing, features, or market positioning), and the system does the rest. The output includes structured evidence tied to sources, a competitor comparison matrix, and a final Markdown report with recommendations. Under the hood, the work is split across five AI agents that each handle one stage of the pipeline. A Planning Agent breaks the question into search queries and quality rules. A Search Agent runs those queries against public sources including Tavily, Exa, Zhihu, and DuckDuckGo. A Fetcher retrieves the actual page content. An Evidence Agent reads that content and extracts specific quotes and facts, recording the source URL, confidence level, and which competitor and dimension each fact relates to. An Analysis Agent then groups the evidence into claims and runs a red-team review step that looks for weaknesses or counter-evidence in those claims. Finally, a Report Agent assembles everything into the final report files. The pipeline includes a coverage check between the search and reporting stages. If the evidence collected is too thin, too uniform in sources, or has too many weak-confidence claims, the system generates additional search queries and loops back to gather more material before proceeding. The backend is built with FastAPI and Python, the frontend is React. LLM calls use an OpenAI-compatible client that can connect to Ark, DeepSeek, or Qwen model providers. All research artifacts are saved locally under a data/runs directory as JSON, Markdown, and CSV files, making individual runs traceable. The project was built for an AI agent competition and is described in the README as a public demo rather than a production multi-tenant product.