On-device computer vision
Fuckable.me
Viral consumer AI without server uploads. Computer vision running directly on your phone.
A high-engagement consumer experiment and active testing ground for client-side computer vision—fine-tuning small vision models to execute entirely within the mobile browser via WebAssembly and WebGPU with zero server data retention.
Visit product
Viral consumer AI usually comes at the expense of user privacy and massive cloud GPU bills.
Most interactive AI products require users to upload intimate selfies to remote cloud servers, creating massive privacy vulnerabilities, data-leak risks, and unsustainable cloud hosting bills for founders. Users are rightly wary of sharing personal photos with opaque third-party databases.
100% client-side inference: private photos never touch a remote server, not even for a millisecond.
We built Fuckable.me as a proving ground for privacy-first edge AI. Lightweight, quantized vision models run directly in the user's browser using WebAssembly and WebGPU acceleration. The entire analysis happens on the device hardware, delivering instant results while guaranteeing mathematically that zero user media ever leaves the phone.
What changed
- Zero server data exposure: 100% of image decoding, feature extraction, and model inference occurs on-device
- Zero cloud GPU costs: edge execution eliminates server-side inference bills, enabling viral scale without infrastructure burn
- Sub-150ms inference latency: hardware-accelerated WebGPU/WASM pipeline delivers instant interactive responses
- Mobile-optimized small vision models: highly quantized neural architectures fine-tuned specifically for smartphone browser engines
- Active architectural testbed: serves as a live proving ground evolving into a broader suite of on-device privacy technologies
- Organic viral loop: shareable card generation that drives distribution without capturing or weaponizing user biometric data
Lightning-fast, viral, and playfully candid—with an invisible fortress of privacy.
The interaction is frictionless: snap or upload a photo, receive an immediate, witty assessment, and share the score preview in seconds. Because processing is local, latency is near zero and users enjoy the experience with complete confidence that their photos remain strictly private.
Certo engineered the on-device inference pipeline: model quantization, WebAssembly compilation, WebGPU compute shaders, browser memory management, and fine-tuning small vision models to run reliably across diverse mobile chipsets.
- Client-side verification: network inspection confirms zero image payloads or binary uploads are transmitted over HTTP/WebSocket.
- Lightweight model footprint: quantized vision architectures optimized to under 15MB for instant mobile web loading.
- Cross-platform acceleration: seamless fallback across WebGPU, WASM SIMD, and WebGL across iOS Safari and Android Chrome.
- Live viral consumer experiment validating consumer adoption and on-device model ergonomics at scale.
Want to build viral AI experiences or sensitive computer vision without cloud GPU bills and privacy liability?
Share your vision pipeline, model requirements, or consumer concept. We engineer ultra-fast, on-device AI architectures that run directly on consumer hardware with bulletproof privacy.