Fraud Detection
Cambium's Marvad AI examination platform had to detect cheating and plagiarism across tens of thousands of papers, but vector-search latency reached 55 minutes and false positives surged as production volumes approached 15 million vectors.
Reported impact
After tuning Atlas search nodes and vector-search settings, Cambium cut similarity-search latency from up to 55 minutes to about three minutes and got Marvad AI live in time for a national rollout.
Cambium used MongoDB Atlas as the data platform for Marvad AI, a national examination system that scans uploaded exam papers and flags cheating or plagiarism through vector-based similarity search. As production data volumes climbed toward tens of thousands of documents and roughly 15 million vectors, search latency ballooned to 55 minutes and false positives threatened the launch. Working with MongoDB Professional Services, Cambium tuned index sizing, memory, and search-node configuration instead of simply adding servers; the team vertically scaled to two S60 search nodes with 64 GB RAM each and reset vector-search parameters to improve precision. The result was a platform that moved from nearly an hour per query to minutes, restored real-time viability, and gave Cambium repeatable guidance for future large-vector AI projects.
Scanned exam papers are uploaded to Marvad AI, embedded into vectors, and stored in MongoDB Atlas for similarity search. Atlas search nodes compare new submissions against the existing corpus, and tuned vector-search parameters drive the alerting workflow that flags likely cheating or plagiarism to the exam platform.
Marvad AI's job is to then check each one for duplications, and trigger an alert if a close match is detected
Overnight, we jumped from waiting almost an hour for a search response to having one come in three minutes.
We scaled vertically to larger search nodes with more RAM
around 15 million vectors could take up to 55 minutes
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
Databricks Sample Apr 23, 2026 Elliptic Elliptic's clients—compliance teams at crypto exchanges, banks, payment providers, and institutional custody platforms—face rising alert volumes and stringent regulation, and analysts must document each decision by… 2x faster resolution of high-risk screenings Agent Bricks / Mosaic AI · Databricks Foundation Model APIs · Databricks Model Serving 
Tencent Cloud Sample Archived · Mar 11, 2026 Aye Solutions Aye needed to prove that face-based payments could resist photos, videos, deepfakes, masks, and replay attacks while still delivering a smooth checkout experience that merchants and end users would trust. Aye has worked with Tencent Cloud for about 2 years. AyeFace · Identity Verification · Liveness Detection 
OpenAI Sample Oct 28, 2025 Doppel Doppel needed to detect, classify, and stop phishing, impersonation, and other social-engineering threats at internet speed, long before human analysts could keep up with the volume and variation created by generative… 3x threat handling capacity GPT-5 · LLM-driven pipeline · o4-mini 
