Architected low-latency, real-time voice AI systems and frameworks for autonomous agent collaboration and task execution.

Optimized high-throughput telemetry using ClickHouse for real-time feedback loops; streamlined delivery via AI-generated code under strict human-reviewed architecture standards.

Integrated GraphRAG to enhance context accuracy and response density across agentic pipelines.

Engineered real-time streaming voice responses over SSE and WebSockets; built multi-turn context-aware NLP for high-performance interactions.

Designed LLM evaluation frameworks for automated output scoring and regression testing; implemented prompt versioning and model routing to reduce inference costs.

Built AI safety guardrails — PII redaction, content filtering, and anomaly detection — across production agentic pipelines.

Architected, designed, and implemented the world’s first edge delivery engine for e-commerce. This included speed performance as well as edge A/B testing, real-time personalization and data engineering to support the platform. Based on AWS (RDS, Redshift, Fargate, Lambda, DynamoDB) and Cloudflare (Workers, R2, and CDN/Load Balancing). Built initial MVP with a team of 3 in just 2 months and then scaled revenue to 2.5MM ARR 10 months later.

✏️ Project Highlights - Through infrastructure-as-code (IAC) implemented and operated the cloud infrastructure and software for a terabyte-scale multi-master redis cluster for the in-memory global edge caches. Managed highly-available container clusters with thousands of container instances for scalable data mining, and edge accelerating content. Implemented petabyte-scale data-service pipelines into both S3 and a Redshift. Managed the transactional data modeling for domain services. Developed multiple ETL workflows, spanning end-to-end from edge capture to data visualization. Developed python notebooks for BI and for data visualization for our customers. Collaborated with statistics PhD to develop and release predictive models (Bayesian). Improved reliability through proactive monitoring and alerting.

Implemented a multi-tenant, cloud-native platform for the existing on-premise radiotherapy application. Refactored a monolith .NET web application to cloud microservices deployed on Kubernetes (EKS) using infrastructure-as-code (Pulumi) through GitHub Actions and Ansible. Worked with medical physicists and mechanical engineers to develop CUDA models to assist in evaluating 3D cancer imaging for anomalies such as over-radiation.

✏️ Project Highlights - containerization of GPU workloads allowed for hybrid deployments onto on-prem GPUs as well as AWS EKS for new SaaS offering and elastic GPU capacity for cost-effective scaling. CICD workflows to create ephemeral development and testing environments, and to validate model updates to FDA compliance through continuous testing. Developed governance-as-code tools in GitHub workflows. Integrated telemetry collectors with time-series Prometheus for Grafana ops dashboards.

Developed full-stack features (Java/Typescript) for the checkout on the e-commerce platform of the Williams-Sonoma brands (Pottery Barn, Mark & Graham, West Elm).

✏️ Project Highlights - Lead the frontend redesign to a single-page-app, and integration into analytics system, where A/B testing showed the acceleration resulted in a 10% increase in revenue for a multi-billion dollar checkout.