Ayush Gupta

• Architectural Foundation: Engineered multi-tenant Next.js environment utilizing Parallel & Intercepting Routes to deliver seamless navigation scaling to 500k+ MAU with optimized SSR/caching delivery.
• Real-Time Video Coordination & Resilience: Designed and maintained the Vooz Video Coordinator state machine managing WebRTC handshakes and Media Recovery Logic for hot-swapping hardware mid-call; established a SignalR Hub Synchronization protocol recovering user state from abnormal disconnections (Code 1006) for 99.9% session continuity.
• Group Audio Platform (0→1): Shipped Hangouts, host-owned group audio rooms on a P2P WebRTC mesh, holding all live room state in Redis behind nine atomic Lua scripts; time-scored membership makes crashed clients and half-open sockets self-expire with no reconciliation job.
• Service Decomposition: Split a monolithic .NET API into independently scaled REST and Realtime services behind one ingress, decoupling WebSocket connection density from REST capacity so hub headroom no longer forces redundant REST pods.
• Database Performance: Resolved production PostgreSQL CPU saturation (~97% → nominal) by rewriting the six hottest queries by pg_stat_statements share — OR→UNION decomposition, covering indexes built CONCURRENTLY, and a username cache eliminating 443M redundant lookups.
• AI Inference at Scale: Spearheaded migration to high-throughput Triton Inference Server & vLLM cluster with Dynamic Batching achieving 1280+ RPS and full OpenTelemetry (OTel) tracing.
• Multi-Rail Payment Infrastructure: Architected payments across crypto and fiat — Solana (Node.js microservice, Helius Webhooks, atomic confirmation) alongside CCBill and Square (Web Payments SDK, 3-D Secure buyer verification, Apple Pay) behind one provider-switchable gateway.
• LLM Orchestration: Integrated LLM APIs (Anthropic Claude, OpenAI) to power conversational features, automated moderation, and semantic search queries.
• Infrastructure & Edge Logic: Orchestrated multi-cloud deployments (Azure AKS / DigitalOcean) secured by ArgoCD (GitOps) and Azure Front Door (AFD) with cookie-based edge routing.
• Client-Side Intelligence & AI Safety: Engineered browser-local inference engine (ONNX Runtime, WebGPU/WASM in a Web Worker) scanning remote video feeds in real time, delivered via the Cache API; paired with backend NSFW services in a dual-track pipeline for a unified blur/report response.

• Led a development team of 3 building AI-powered mobile applications using React Native (Expo).
• Designed & deployed scalable RAG & Multimodal AI systems integrating Claude, OpenAI GPT, LLaMA, image generation, and open-source AI models.
• Developed backend orchestration services in Node.js & Python for semantic retrieval workflows and LLM caching.
• Authored custom native modules in Swift & Java bridging native mobile features with AI context layers.

• Optimized critical React Native application flows for high traffic, eliminating memory leaks and improving frame rates.
• Collaborated with backend teams to strictly type REST & GraphQL API contracts for cross-platform data consistency.

High-throughput GPU inference backend powering 1280+ RPS across production LLM and vision workloads. Engineered with Dynamic Batching, Prometheus metrics, and OpenTelemetry distributed tracing.

Browser-native inference engine evaluating real-time video feeds on-device using WebGPU execution providers and Cache API, eliminating cloud inference overhead.

Trained deep learning models using TensorFlow on PubMed RCT datasets. Achieved 90%+ accuracy in classifying unstructured medical research abstracts.

Cross-platform signaling and WebRTC state machine featuring end-to-end encryption, Media Recovery logic, and SignalR hub sync for 99.9% session continuity.