An AI due-diligence agent that cannot fabricate a quote.

IT, OT and cybersecurity vendor reviews are long manual work: reading every response, building the matrix, cross-referencing sections and drafting questions. Xinc asked whether agents could do the reading and the first draft, and leave the judgment to people.

Reneka built a six-stage pipeline. One agent turns each requirement into the pass bar it states. Others rate each vendor and quote its own words. Plain code then checks every quote word for word against the source, so a fabricated quote cannot pass. A cross-check agent reads the whole response for contradictions, costs outside the headline price and skipped requirements. A skeptical second agent re-rates anything uncertain, and the memo is written only from verified results.

The test was a fictional sample: a pipeline operator choosing a managed OT and IT security provider, with 26 requirements, three vendor responses and deliberate problems recorded in an answer key. In the final run the agent caught 30 of 30 planted problems and 3 of 3 cross-section contradictions, verified 116 of 116 quotes, and finished in under two minutes of machine time.

It also over-flagged 9 of 33 clean answers as partial, which is why a person reviews every result. This is a prototype result on synthetic data. Time saved on a real review has not been measured yet.