Namaste

• Completed: Implemented a speech-to-text pipeline using Deepgram, enabling low-latency and high-accuracy voice transcription.
• Built a modular Python + Flask backend, supporting scalable integration of multiple translation engines.
• Evaluated and benchmarked ASR models (Whisper, Gladia, ElevenLabs Scribe) based on accuracy, latency, and system constraints.
• Roadmap: Integrate Azure Translator to enable multilingual text translation with automatic language detection.
• Implement real-time WebSocket streaming for live transcription and translation delivery.
• Deploy the application using Docker and cloud hosting, with added performance metrics and reliability controls.

ZERO BALANCE is an online platform designed to load quickly and work reliably for users across different internet networks. It focuses on smooth performance, secure access, and a clean user experience, and is built in a way that allows new features to be added easily in the future.

• Built a cloud-based web app using Vercel, Supabase, and AWS S3 with CDN-backed static asset delivery.
• Configured custom domains, SSL, CDN caching, and secure DB access, reducing page load time by 45%.
• Designed scalable VPS based frontend–backend architecture supporting future auth, APIs, and analytics.

• Developed a full-stack access-controlled platform using Vercel and Railway with Cloudflare DNS and SSL.
• Implemented mobile-number–based authentication and authorization, reducing unauthorized access by 90%.
• Built rule-based user routing to assign WhatsApp groups dynamically, cutting onboarding failures by 60%.

• Developed an interactive VR-based surgical training system simulating hernia repair with real-time controls.
• Reduced dependency on physical training resources, lowering training costs by 30%.

• Built a face-detection system for crowd tracking that made counting 40% faster and more reliable than manual methods.
• Enabled real-time updates (latency <200ms) via Firebase integration.

• Processed video feeds at 25–30 FPS ensuring low-latency responsiveness.
• Achieved ~95% detection accuracy using EAR-based metrics in real-time monitoring.