Solutions & products.
Two are in production today. Two are pilot-ready. One is in regulator engagement. Access to architecture documents and live demos is gated to verified prospects, partners, and investors.
Catalogue
Each product is designed to be licensed, deployed, and adapted. Status pills indicate maturity; "production" means the system is running real workloads today.
National gambling harm-prevention infrastructure. Cross-operator behavioural detection across a four-layer architecture (registry → analytics → intervention → regulatory intelligence). No domestic competitor in the SA R75 billion market.
Market
R75B
Operator-side bettor analytics SaaS, the commercial twin of Responsible-BAE. PAS, PPS, BIF, AdjustedSER, and BVI as licensable APIs for retention, risk, and value scoring.
Throughput
>10K/day
Autonomous financial defence for credit-distressed consumers. NCA-aligned, advocacy-on-behalf, designed for the realities of South African household debt.
Status
Legal review
AI-based road-condition assessment and fleet intelligence. Pothole detection, accident classification, and ANPR, trained on a 62 GB proprietary fleet-video corpus (PRIMUS_DATA).
Data moat
62 GB
Three model classes, one inference graph: YOLO26 for hazard detection (NMS-free end-to-end; ProgLoss + STAL assignment for small-object recall, since a 5 cm pothole at 30 m is a small object); SAM 3 for severity classification (text-to-mask isolates damage; mask-area + edge-sharpness regression produces a four-class severity score aligned to SANRAL inspection bands); YOLO26 + PARSeq OCR for ANPR, validated against SA plate formats.
62 GB of fleet dashcam at 720p to 1080p. Keyframe-anchored sampling (1 fps under 60 km/h, 2 fps above), hard-negative mining around driver-flagged false positives. Pseudo-mask annotation via SAM 3 text-grounding seeded by 4,000 hand-annotated reference frames; humans verify edge cases.
Edge-first. Inference runs on the fleet vehicle's NPU, and only structured detections and plate hashes leave the device. Raw video stays on-vehicle unless retention policy triggers upload (POPIA §19 minimum-data principle). Cloud aggregator runs in SA-resident regions; cross-border export disabled by default. RAF 2025/00013-aligned variants emit eCrash-format incident packets for direct claim ingestion.
Self-hosted AI workspace for institutions that cannot send their data to the public cloud or to foreign LLM endpoints. Runs on customer infrastructure, governed by customer policy.
Deployment
On-prem
Verified prospects, partners, and investors only. We typically grant access within 24 hours of NDA signature.
The IP behind the products
All five products draw on the same metric library, the mathematical moat under the Braigns platform. Definitions and methodology are detailed in the Responsible-Gambling white paper (NDA-gated).
Player Action Score. Behavioural risk for harm-detection and intervention prioritisation.
Player Profitability Score. Commercial value for retention and lifecycle optimisation.
Behavioural Intensity Factor. Temporal pattern weighting for short-window risk.
Adjusted Skill-Expected Return. True-skill estimation under stake variance.
Bettor Volatility Index. Variance signature, predictive of distress states.
Profit Influence Index. Operator-level attribution of bettor cohorts to revenue.