Gravicus,
turning data into signal.
ML and NLP that turn sprawling messy datasets into compliance, risk and business signal. I led the visualisation work, the SAR relationship maps and the wireframing of an investigator's workflow that didn't bury them in entities.
Role
Senior Product Designer
When
2018 - 2019
Team
2 PMs, data science, 6 engineers
Product
Osprey · SAR, DPIA, Privacy, BackOffice
Background
Gravicus built Osprey, an ML and NLP product for large enterprises. The simple version: companies sit on enormous, fragmented document stores. Auditors, compliance officers, legal teams need to find specific signals inside that mass. Privacy breaches, conflicts of interest, sanction risk, data subject requests. Osprey crawled the data, surfaced the relationships and gave investigators a workbench.
The data science was strong. The interface was buried. I came in to make the model output usable by a non technical analyst sitting in compliance at a bank.
UX Design
Connecting data sets
Setup flow for plugging in document stores. SharePoint, network drive, mail archive, ticketing. The UX is mostly about giving the user faith their corpus is properly indexed.
Collections
Custom slices based on parameters that begin real time analysis of their data. Less "saved search", more "live working set".
Data alerts & NLP
The system behaves like an anti virus scanner. Alerts on potential data risk. UX is about making the alert specific enough that the analyst doesn't have to triage 500.
SAR workflow
A Subject Access Request is a regulated process. Walk the analyst from a name through every linked document. Search, redact, package, send.
SAR · entity graph and relationship map
Collections & alerts · the analyst's daily surface
UI Design
Visualisation library
Chose React Vis over D3 for the production library. Easier maintenance, better animation defaults, less custom glue.
Entity graph
The relationship map sits at the heart of every investigation. The discipline was not adding everything. It was choosing what to hide by default.
Risk scoring surface
Confidence and risk are different. We surfaced both. The investigator should always be able to tell whether the model is sure but wrong, unsure but right or confidently flagging the obvious.
Across five surfaces
SAR, DPIA, Privacy, BackOffice and Dashboard. Each with its own working idioms but sharing the same primitives.
Outcome
Osprey shipped across five product surfaces. Enterprise customers integrated the platform into their compliance and privacy operations. The pattern of "show the relationships, hide the documents until asked" became the way the whole product worked. Gravicus went on to be acquired by Smartbox.
Zing.gg→
Payments platform with KYC and AML in a single API. Merchant onboarding and agent back office.