AuthentiFlow

🚀 AuthentiFlow: Continuous Behavioral Authentication for Mobile Banking 🧠 About the Project 💡 What Inspired Us In today’s fast-moving digital world, passwords and even biometric logins offer only point-in-time security. Once a user logs in, apps typically assume the session remains secure — even if the phone is passed to someone else, stolen, or compromised.

We wanted to challenge this assumption and build a smarter system: one that verifies users continuously based on how they interact with their device, not just when they log in. Our idea, AuthentiFlow, is a mobile banking security app that continuously authenticates users using their behavioral biometrics — like typing patterns, swipe gestures, and device motion.

Imagine your phone noticing it’s not you holding it — even after a successful login. That’s the future we’re building.

🛠️ How We Built It We developed AuthentiFlow as a mobile SDK that can be embedded in any banking app. It runs silently in the background after the user logs in and continuously tracks behavioral signals using on-device sensors.

Core Technologies Kotlin (Android) for app development

SensorManager and MotionEvent for capturing behavior (typing, swiping, etc.)

TensorFlow Lite for real-time, on-device anomaly detection

EncryptedSharedPreferences for secure data storage

One-Class SVM & Isolation Forests for behavior modeling

Exponential Decay Model to keep updating the user’s baseline without constant retraining

Behavioral Vectors Captured Typing speed and keypress intervals

Tap pressure and duration

Swipe direction, speed, and frequency

Screen navigation flow

Device orientation and motion

(Optional) GPS pattern for high-risk transactions

Privacy by Design All processing is done offline on-device

Users can opt-in/out of any behavioral tracking

Data never leaves the phone without user consent

🤯 Challenges We Faced Sensor Noise and Inconsistency

Different phones have different sensor accuracy and sampling rates.

Solved by smoothing input data and creating phone-specific calibration layers.

False Positives in Anomaly Detection

Not all behavior changes are suspicious — a tired user types differently!

Introduced a multi-stage threshold system: warn first, then block only if behavior remains erratic.

Power Efficiency

Continuous tracking can drain battery if not optimized.

We implemented event-driven data collection instead of polling.

Privacy Concerns

Behavioral biometrics are sensitive. We strictly adhered to local processing and anonymization, ensuring no raw data is stored or sent online.

📚 What We Learned Behavioral biometrics offer a rich and largely untapped source of authentication signals.

Real-time, on-device machine learning is very much feasible with today’s tools.

Users value security as much as they value transparency and control — privacy-first design is non-negotiable.

Building such systems requires a delicate balance between security, usability, and energy efficiency.

💬 What’s Next We plan to:

Expand to iOS with Core ML and Swift.

Add voice behavior profiling for enhanced multimodal authentication.

Create a full dashboard for banks to customize security policies.

Open-source a lite version for the developer community.

🏁 Final Thoughts AuthentiFlow goes beyond passwords and biometrics. It's a step toward continuous, intelligent, and privacy-first mobile security. We believe this approach could revolutionize not only banking but any mobile app where trust matters.

Security isn’t a moment — it’s a flow. With AuthentiFlow, that flow is always protected. 🔐

Built With

• ai
• bolt
• chatgpt