Mobile Face Verification
Budget: ₹600 – ₹1,500 INR
I need a lightweight face-verification solution for our field workers that runs entirely on their mobile phones, with no dependency on a data connection once the model is installed.
Here is the workflow I have in mind: each worker’s reference image is captured during onboarding, stored securely on the device, and whenever the app is opened it quickly checks the live selfie against that stored template. A simple “verified / not verified” result is all that is required on-screen, plus a local log so the outcome can be synced to our server whenever connectivity is restored.
Key expectations
• Works offline after initial installation and enrolment
• Fast verification (under two seconds on mid-range Android)
• Local biometric templates encrypted at rest
• False-accept and false-reject rates comparable to commercial SDKs (we can fine-tune together)
• Clean, well-commented source code so our internal team can maintain it
Preferred stack is Kotlin or Flutter paired with TensorFlow-Lite or any edge-optimized library you are comfortable with; I’m open to other suggestions if performance is better. If you already have an engine that can be re-skinned, let me know.
When you reply, please confirm:
1. Which on-device model or library you propose
2. Hardware you will use for performance benchmarks
3. Typical FAR/FRR you can achieve in offline mode
A quick prototype APK is the first milestone; full source and brief documentation will close the project.
Here is the workflow I have in mind: each worker’s reference image is captured during onboarding, stored securely on the device, and whenever the app is opened it quickly checks the live selfie against that stored template. A simple “verified / not verified” result is all that is required on-screen, plus a local log so the outcome can be synced to our server whenever connectivity is restored.
Key expectations
• Works offline after initial installation and enrolment
• Fast verification (under two seconds on mid-range Android)
• Local biometric templates encrypted at rest
• False-accept and false-reject rates comparable to commercial SDKs (we can fine-tune together)
• Clean, well-commented source code so our internal team can maintain it
Preferred stack is Kotlin or Flutter paired with TensorFlow-Lite or any edge-optimized library you are comfortable with; I’m open to other suggestions if performance is better. If you already have an engine that can be re-skinned, let me know.
When you reply, please confirm:
1. Which on-device model or library you propose
2. Hardware you will use for performance benchmarks
3. Typical FAR/FRR you can achieve in offline mode
A quick prototype APK is the first milestone; full source and brief documentation will close the project.