Series-B freight platform
Customs document intake, from six minutes to fourteen seconds
- Handling time per shipment
- 6 min → 14sHandling time per shipment
- Documents processed monthly
- 340k/moDocuments processed monthly
Eleven sites, each with its own accumulated clinical guidance, some of it contradicting the others. Clinicians were spending measurable time hunting for the current version of a protocol. Any solution that sent a single patient identifier to a third-party API was categorically out.
Sutra deployed inside the network's own VPC with self-hosted inference. Hybrid retrieval over the guidance corpus, with every answer citing the source document, the site it applies to and the revision date. Where two sites disagree, the assistant says so and shows both rather than silently picking one. An evaluation set written with the clinical governance team ran in CI from the first week.
Clinician research time down 71% against a pre-deployment baseline the governance team defined themselves. No patient data left the network at any point — there was no egress path to leave by. The contradiction-surfacing behaviour turned out to be the feature clinical governance valued most, which nobody predicted during scoping.
Series-B freight platform
Challenger bank, 400k customers
Send the problem, not a spec. We'll come back within one business day with an honest read on whether we're the right team — and who to talk to if we're not.
attatvasystems@gmail.com · Replies within 1 business day, Kathmandu time (UTC+05:45)