Heysamay
HEYSAMAY is an officially registered technology enterprise under the Ministry of MSME, Government of India (Udyam Reg. No: UDYAM-RJ-23-0028745), engineering real-time speech translation terminals, Heysamay Sambandh speed blind dating & social audio rooms, intelligent dietary tracking, and privacy-first document tools.
Scan your meal plate using computer vision AI to instantly get calorie counts, macronutrients and nutritional data. Also features real-time on-device voice translation across 80+ languages using quantized ONNX neural models — works offline with complete privacy.
Host or join real-time multilingual translation rooms where everyone speaks in their own language and hears others translated instantly. Powered by distributed WebSocket synchronization and on-device ONNX speech models.
A complete online PDF and document utility platform. Merge, split, compress, convert, sign, watermark, OCR, and AI-translate documents. 100% client-side processing — files are never uploaded to any third-party server.
Explore our technical knowledge base covering edge artificial intelligence, computer vision in dietetics, and acoustic speech engineering:
Architectural breakdown of quantized 8-bit transformer models running on mobile ARM64 hardware.
How deep learning models segment meal items, estimate 3D volume, and compute macronutrients.
Decentralized WebRTC signaling and real-time fan-out pipelines for global conference collaboration.
Preventing speaker-to-microphone acoustic loopbacks during single-phone shared translation.
A science-backed guide to energy balance, protein synthesis, carbohydrates, lipids, and vitamins.
How compiled Rust Wasm and WebGPU kernels enable browser-side PDF editing and OCR without server uploads.
Analyzing systemic vulnerabilities of cloud health data lakes and local mobile edge AI safeguards.
Monocular depth estimation, RANSAC plane fitting, and Delaunay 3D volumetric integration for meal analysis.
Designing inclusive translation terminal hardware under DSIR PRISM guidelines for high-noise windows.
Yes. By downloading our regional ONNX neural model packages, translation inference executes entirely on your phone's processor without sending audio to the cloud.
The system provides nutritional approximations based on volumetric segmentation and validated composition tables. For complex home-cooked dishes, users can provide optional voice descriptions to refine calorie estimates.
No. Speech audio processed on-device is never recorded or uploaded. Translation room streams are ephemeral and purged upon room termination. Diet scan photos are stored privately under your user account and can be deleted at any time.