Project Overview: Built and operated generative-AI backend services for multimodal content (text-to-image, image-to-video, audio) running at consumer scale.

• Designed production inference services that handled 1M+ monthly interactions, exposing standardized REST endpoints and SDKs for web/mobile integration. • Implemented temporal attention and motion-aware modules for video generation; reduced artifacts and improved output realism across core use cases. • Built GPU-optimized, containerized pipelines on Kubernetes with autoscaling and request batching; cut p95 latency by 38% under peak load. • Applied model compression (quantization, pruning) and precision tuning to lower GPU hours while maintaining quality and throughput targets. • Stood up end-to-end observability with Prometheus/Grafana, structured logging, and alerting tied to SLOs; enabled proactive capacity tuning and rapid incident response. • Established auth, rate limiting, and usage metering for multi-tenant clients; aligned IAM policies and service boundaries with least-privilege access.

Project Overview: Delivered conversational AI services that integrated with scheduling, billing, and membership systems to streamline member support

• Architected a chatbot backend using Django APIs and background workers to orchestrate intents, tool usage, and secure data retrieval from core platforms. • Trained and fine-tuned NLU models (BERT/TensorFlow) with domain corpora; increased intent coverage and contextual accuracy for fitness workflows. • Built a curated knowledge base and retrieval layer to standardize answers across policies, schedules, and FAQs; improved response consistency. • Introduced analytics pipelines for dialog events, abandonment, and CSAT; informed iterative model and UX improvements that reduced support load. • Implemented monitoring, feature flags, and safe rollouts for model updates; maintained uptime during peak traffic and seasonal campaigns. • Hardened security and compliance with input validation, secret rotation, and IAM scoping across environments.