We partnered with the Setara AI founder across three disciplines. Statistical analysis, model development, and system architecture. To build the platform's core intelligence layer. It is now in production with legal, cyber, risk, and executive teams.
Statistical models delivered
Entity extraction, inference, anomaly detection
Architecture components
Ingestion through output generation
Ingestion to briefing
Counsel-ready output, end to end
Production platform
Live with legal, cyber, risk, and executive teams
The Setara AI founder had a clear vision: a platform that turns scattered evidence across emails, logs, documents, and public signals into a coherent intelligence picture.
The challenge was translating that vision into a model architecture that could deliver it at scale, under real investigative conditions, across high-volume, heterogeneous data sources.
The work spanned the full intelligence stack. From the statistical methods at the base, through the AI models that reconstruct context, to the architecture that runs them in production.
We designed the statistical methods underpinning entity extraction, relationship inference, and anomaly detection: model selection, validation frameworks, signal-to-noise separation, and the probabilistic matching models that connect external OSINT signals to internal entities across name variants, identifiers, and temporal proximity.
We developed the core AI models for context reconstruction: turning fragmented inputs into timelines, evidence maps, and relationship pathways. Iterative prototyping, performance benchmarking, refinement against real investigative scenarios. The OSINT enrichment pipeline was built as a distinct component. Scoring public signals against the internal evidence graph without contaminating chain-of-custody integrity.
We designed the system architecture for ingestion, processing, and output generation. Enterprise security, tenant isolation, evidence integrity, and deployment flexibility were primary constraints, not afterthoughts.
Most investigation tools are bounded by what happened inside the organisation. Logs, emails, files, chats. That boundary is their limitation.
We built the OSINT enrichment layer that crosses it. Setara cross-references internal evidence against public signals: corporate filings, domain registrations, company formations, professional histories, court records, public announcements, and open-source intelligence feeds. Normalised in real time against the internal evidence already in the system.
Corporate filings, domain records, professional profiles, court records, open-source intelligence feeds.
External signals cleaned, deduplicated, and matched against internal entities already extracted by the model.
OSINT does not add noise. It adds the external facts that change what internal evidence means.
"An access event becomes more significant when OSINT shows the same person registered a competing entity the week before. Internal facts and external context are no longer separate."
Mapping the investigative context, the evidence fragmentation problem, and the performance requirements. Defining what intelligence actually meant for Setara's users.
Selecting the statistical approach, defining the model architecture, designing the system layers. Evidence integrity, tenant isolation, and the boundary between automated inference and reviewable output.
Iterative development against real-world investigative scenarios. Validation, benchmarking, refinement, and preparation for production deployment.
Setara AI is now a production intelligence platform used by legal, cyber, risk, and executive teams to reconstruct complex events from fragmented evidence.
The platform delivers timelines, evidence maps, relationship pathways, and counsel-ready briefings in under 48 hours.
Evidence resolved into structured intelligence. Indexed
Model, architecture, or both. We scope it on the first call.
Setara AI engagement delivered through the Human Nexus group.
If you have a vision that needs a model, an architecture, or both. We design and build it.