Scout

Manual prospect research is slow, noisy, and easy to repeat. Finding an organization is only the first step: a useful lead needs context, evidence, a plausible problem, and a reason to investigate further. As the geographic search space grows, previously discovered organizations and accepted opportunities also need to influence what gets researched next.

Treat discovery as an evolving research workflow rather than an autonomous sales machine. Scout searches within a geographic and business context, assembles evidence, evaluates a candidate's practical problem and fit, and produces a structured report for review. Previously surfaced pipeline context is passed into the run so the system can account for existing opportunities instead of rediscovering the same obvious leads indefinitely. Human review remains the boundary between a research signal and a working opportunity.

Scout is a configured Python CLI with a linear search → evidence → analysis → buyer gate → technology landscape → qualification → ranking → report pipeline. The repository isolates Tavily and replayed-JSON search providers, uses OpenAI structured outputs behind an LLM provider, validates typed evidence and opportunity models with Pydantic, and writes both a readable Markdown brief and replayable JSON run artifacts. Command Center launches it without a shell, captures its run status and bounded output, finds the newest Markdown report, and parses the report into reviewable candidates. The report carries scores, industry, location, likely buyer, observed problem, evidence, existing-solution risk, Mindjax role, next action, and discovery questions. Imported candidates remain separate until a person adds one to the opportunity pipeline.