Insurance Reimagined by AI
Black Swan Insurance was founded on a simple belief: insurance should be as intelligent as the risks it covers. We built a platform where machine learning meets actuarial science.
Making Protection Personal
Traditional insurance treats everyone the same — blunt risk bands based on age and profession. We believe your health profile is unique, and your coverage should be too.
By applying computer vision and machine learning directly to health indicators visible in a simple selfie, we can price risk fairly, instantly, and at scale.
Why Black Swan?
A "black swan" event is one that defies conventional prediction — rare, impactful, and only obvious in retrospect. Our AI identifies health signals invisible to the naked eye, preparing you for outcomes traditional insurers never see coming.
The AI Platform
Four production ML models working in concert — all running on CPU for cost-effective, scalable inference.
Industry-standard face detection and alignment model. Detects faces, extracts 5-point landmarks, and performs affine alignment — ensuring all downstream models receive a consistently normalised 224×224 crop.
A 26-class softmax regression model trained on paired face–BMI datasets. Outputs a probability distribution across BMI bins (15–40), from which a weighted expectation is computed.
OpenCV DNN module running Caffe's age classification network. Outputs age from 8 bins (0–2, 4–6, 8–12 … 60–100) with calibrated midpoints for continuous age estimation.
EfficientNet-based classifier exported to ONNX for fast, portable CPU inference. Achieves >95% binary classification accuracy on balanced test sets.
A two-stage pipeline: face parsing removes background clutter (BiSeNet), then a ResNet binary classifier detects chronic tobacco-use skin signals. Background subtraction reduces false positives by ~30%.
The backend uses FastAPI with asyncio.gather() to run all four models concurrently in thread pools, keeping total latency close to the slowest single model.
Your Data, Protected
All data in transit is encrypted with TLS 1.3. Self-signed certificates for development; Let's Encrypt in production.
Images are processed in-memory. Temporary files written during analysis are purged immediately after the response is sent.
Biometric processing follows EU GDPR Art. 9 guidance. No facial embeddings are stored or shared with third parties.
All endpoints are rate-limited (100 req / 15 min) with Helmet.js security headers and Content Security Policy enforcement.
Our AI models are documented using NIST AI Risk Management Framework standards — including model cards, bias audits, and governance policies.
CSP, HSTS, X-Frame-Options, and other security headers enforced via Helmet.js on the Node.js layer.
How It All Connects
Ready to Experience AI Insurance?
Join thousands of customers who have already discovered their personalised insurance profile. It takes less than 10 seconds.