Check a company for selected risk topics and return the sources for each finding.
Adverse-media screening needs recent reporting and a consistent way to separate relevant incidents from unrelated mentions. Exa searches each risk topic independently and returns the incidents, severity, and source pages together for analyst review.
This is a live workflow. For every selected risk topic, the demo runs up to two independent Exa Deep Searches by default, then combines the topic-level findings into the company screen.
The user selects built-in topics such as fraud, sanctions, litigation, and regulatory action, with optional custom topics. Keeping each topic separate prevents evidence for one risk from leaking into another verdict.
Each sample sends POST /search with type deep, numResults 25, a topic-specific system prompt, and a schema requiring incidents, severity, and summary. SCREEN_SAMPLES defaults to two and is configurable from one to five; sampling stops as soon as qualifying evidence is found. Calls use bounded concurrency and retry transient throttling or server errors.
/search
A topic is flagged only when at least one returned incident is both in scope and about the screened company. Flagged topics use citations from output.grounding; cleared topics show up to six top-level search results that Exa considered. The response also computes the highest topic severity for its overallRisk field.
The interface presents surfaced coverage rather than an automated verdict. Sources remain attached to the topic and sample that produced them.
The result is a review aid, not an automated legal conclusion. A cleared topic means the configured search did not return qualifying evidence; analysts should still apply their own policy, identity checks, and escalation rules.