VAISHAKH I KUPPAST

Architected a fully serverless matrimonial platform for the Shivasimpi community — running entirely on free-tier infrastructure. Built with GitHub CDN for static hosting, Google AppScripts for backend logic, Firebase for real-time data, and Mail APIs for notifications. The platform serves 1000+ active members with profile matching, community-specific filters, and admin tools — all at zero operating cost. What started as a freelance project at age 18 evolved into a sustainable community service that runs itself.

Designed and built an AI-powered Slack chatbot that became the go-to knowledge assistant for 2000+ internal engineers at AWS. Integrated Amazon Lex V2 for natural language understanding, Kendra for intelligent document search across internal wikis and runbooks, and Bedrock for generative AI responses. The bot handles everything from troubleshooting queries to onboarding questions — reducing query resolution time by 80% and saving the team 20+ hours every week. Went from prototype to org-wide adoption in under 3 months.

Authored 20+ simulation-based challenge labs spanning 50+ AWS services — designed to teach engineers by doing, not reading. Each lab drops users into a realistic broken environment and challenges them to diagnose and fix issues using real AWS consoles and CLIs. Covers everything from VPC networking and IAM policies to CloudFront caching and Lambda debugging. The program achieved a 90% completion rate and 4.8/5 satisfaction score, reducing new hire onboarding time by 40%. Now used as a standard training resource across the organization.

Built an automated migration tool that converts CloudFlare CDN configurations to equivalent AWS CloudFront distributions. The tool parses CloudFlare zone settings, page rules, WAF rules, and DNS records — then programmatically creates matching CloudFront distributions, behaviors, Lambda@Edge functions, and Route 53 entries using AWS SDK. What used to take engineers 5 days of manual configuration and testing now completes in 2 hours with 90% fewer manual steps and built-in validation checks.

Core research project behind Aura VR Education. Developed a novel method for playing 4D models — reconstructed from live motion capture of humans — on resource-constrained Android devices. Solved the challenge of rendering temporally-varying 3D meshes in real-time AR by implementing progressive mesh streaming, LOD switching, and GPU-optimized rendering pipelines. The technique enables immersive AR experiences on mid-range phones without requiring cloud rendering.

Built a multi-modal emotion recognition system that fuses audio and video inputs for real-time emotion detection. Combined VGG-Net for facial feature extraction with LSTM networks for temporal sequence modeling across video frames, plus a parallel audio emotion classifier. The three models are fused using a weighted decision layer optimized for on-device inference. Deployed on Android with TensorFlow Lite — achieving real-time performance on mid-range devices.

Tackled the challenge of detecting and recognizing text appearing in arbitrary orientations in natural scene images — signboards, labels, documents at angles. Built a pipeline using CNN for text region proposal and Caffe FCN for pixel-level text segmentation, followed by orientation correction and OCR. The system handles rotated, curved, and perspective-distorted text that traditional OCR engines fail on.

Built under Samsung's PRISM research program. Designed an end-to-end ML pipeline to automatically filter and categorize crowd-sourced image datasets. The pipeline includes shadow detection using histogram analysis, blur detection via Laplacian variance, occlusion detection using YOLO object detection, and text extraction with Tesseract OCR. Automatically selects high-quality, properly-framed images from noisy real-world data — directly feeding Samsung's computer vision research initiatives.

Industrial research project focused on digitally preserving and visualizing Indian heritage sites through mixed reality. Built an Android application that overlays historically accurate 3D reconstructions of monuments and artifacts onto real-world camera views. Users can walk around physical heritage sites and see them restored to their original glory through their phone screens — complete with contextual information, audio narration, and interactive exploration.

My first real project — built at age 18 as a freelance developer. A matrimonial app for the Aqila Bharatha Shivasimpi Samaj community featuring dynamic face cropping for profile photos, community-specific matching filters, and family-tree integration. The app made local newspaper headlines and hit 100+ downloads on Google Play Store. This project taught me everything about shipping a real product — from user research and UI design to backend architecture and Play Store deployment.