ClearPass AI

Inspiration

Across Africa and many developing regions, financial inclusion is hindered by two massive roadblocks: onboarding friction and rampant digital fraud. For millions of unbanked individuals, jumping through manual identity verification hoops prevents them from accessing basic digital banking. Simultaneously, fintechs are bleeding money due to "ghost borrowers" and coordinated fraud rings. We were inspired to build ClearPass AI to bridge this gap—creating a portable, frictionless identity infrastructure that protects institutions while democratizing access to financial tools.

What it does

ClearPass AI is a high-performance identity verification and risk-scoring platform. Instead of relying on a single data point, it uses a proprietary Triple-Layer Risk Brain to generate a "Multi-Model Trust Score":

1. Biometric Integrity (Vision Layer): Validates the user in real-time using Active Liveness (blink detection) and matches their selfie against government IDs.
2. Behavioral Intelligence (Risk Layer): Analyzes financial "rhythms" (income consistency, debit/credit ratios) to predict borrower reliability and flag outlier spending behaviors.
3. Identity Graph (Network Layer): Maps connected identities (BVNs, devices, locations) to automatically detect and block coordinated fraud rings.

It acts as a "Trust-as-a-Service" gateway, allowing fintech platforms to safely approve legitimate users in milliseconds.

How we built it

We built ClearPass AI as a robust backend platform using Python and SQLite. We integrated several state-of-the-art Machine Learning frameworks to power our Risk Brain:

• Vision Layer: We used MediaPipe FaceMesh to track 468 3D facial landmarks for client-side active liveness (reducing server bottlenecks), and FaceNet embeddings to calculate the Euclidean distance for biometric matching.
• Risk Layer: We implemented an XGBoost profiler to predict borrower reliability and used an Isolation Forest model to hunt for "Ghost Borrowers" showing anomalous financial rhythms.
• Network Layer: We used NetworkX to build the identity graph for clustering and fraud ring detection.
• Transparency: To ensure regulatory compliance, we integrated SHAP (SHapley Additive exPlanations) to provide Explainable AI (XAI), giving human-readable reasons for every trust score generated.

Challenges we ran into

One of the biggest technical hurdles was managing latency. Processing complex computer vision models on a server can cause massive bottlenecks. We solved this by offloading the Active Liveness (FaceMesh) tracking to the client-side, achieving sub-second performance. Additionally, ensuring our ML models were not a "black box" was a challenge, which required us to deeply integrate SHAP to explain exactly why a user was flagged or approved.

Accomplishments that we're proud of

We are incredibly proud of our Explainable AI (XAI) integration. In fintech, you can't just deny a user without a reason. Building a system that not only detects fraud with high accuracy using Isolation Forests, but also generates a transparent, human-readable compliance report via SHAP, makes this a truly production-ready tool. We're also proud of our "Verify Once, Use Everywhere" reusable identity model.

What we learned

We gained deep, hands-on experience orchestrating multiple distinct ML models (Computer Vision + Tabular ML + Graph Networks) into a single, cohesive pipeline. We also learned how to balance security with user experience—realizing that pushing computer vision tasks to the client-side drastically improves the onboarding flow for users in regions with slower internet connections.

What's next for ClearPass AI

We plan to expand the platform by:

Alternative Data Scoring: Integrating mobile money transaction histories and telco data into the XGBoost model to better score unbanked individuals who lack traditional banking records. Decentralized Identity: Moving the core identity tokens to a blockchain ledger to give users complete ownership over their biometric and financial data. Edge ML: Further optimizing our FaceNet models to run entirely on low-end Android devices, making the onboarding process truly offline-first.

Built With

• artificial-intelligence
• computer-vision
• explainable-ai
• fintech
• machine-learning
• mediapipe
• networkx
• python
• social-good
• sqlite
• xgboost