The problem and intended users
Search and AI-discovery analysis can become misleading when different kinds of evidence are treated as the same measurement. SearchIntel-ai is intended for people assessing how information is represented in model knowledge, controlled web search and answers grounded in a first-party site.
Three distinct measurement modes
The repository separates memory, web_search and site_rag. Memory concerns latent model knowledge; web_search uses controlled API search; site_rag examines answerability against crawled first-party content. The distinction is important: a grounded site answer is not proof of an external search citation, and the three modes are not quantitatively interchangeable.
Documented architecture
The application is described as a FastAPI backend with SQLAlchemy, PostgreSQL and Alembic, alongside a Next.js dashboard. No measured visibility improvement, ranking movement or customer outcome is asserted.
Limitations and interpretation
The V1 deployment notes identify agency-wide staff permissions and in-process background benchmark execution. AI execution depends on configured provider access; being able to start the application does not prove a provider run will succeed. The engineering lesson is to label the source and limits of each observation before combining results.