Things I've built.
Agentic AI systems, full-stack platforms, and the backend infrastructure that keeps them running in production.
Architected a scalable ads platform using Java Spring Boot microservices and domain-driven data modeling on PostgreSQL, onboarding thousands of non-technical entrepreneurs with sub-10-minute campaign launches. Integrated the Facebook Marketing API and Mapbox for geographic targeting, cutting campaign setup errors by 70% and removing the need for Meta Ads Manager entirely.
Built an n8n workflow orchestrating 5–7 OpenAI agents with AutoGen and prompt engineering to automate ad copy and creative generation, including text-to-image generation, cutting production time from hours to about 2 minutes per campaign.
Built multimodal, voice-enabled AI learning assistants handling student onboarding, live lecture transcription, and conversational tutoring using Whisper, Deepgram, and LiveKit. Backed by multi-threaded FastAPI + WebSocket services supporting concurrent real-time audio sessions in production classrooms.
Designed vector database architectures and retrieval-augmented generation (RAG) systems handling large-scale multimodal data and embeddings, deployed on AWS SageMaker and Bedrock via Terraform, cutting inference costs 40% while scaling to millions of monthly requests — all while keeping student data FERPA-compliant.
Things I've built.
Agentic AI systems, full-stack platforms, and the backend infrastructure that keeps them running in production.
Architected a scalable ads platform using Java Spring Boot microservices and domain-driven data modeling on PostgreSQL, onboarding thousands of non-technical entrepreneurs with sub-10-minute campaign launches. Integrated the Facebook Marketing API and Mapbox for geographic targeting, cutting campaign setup errors by 70% and removing the need for Meta Ads Manager entirely.
Built an n8n workflow orchestrating 5–7 OpenAI agents with AutoGen and prompt engineering to automate ad copy and creative generation, including text-to-image generation, cutting production time from hours to about 2 minutes per campaign.
Built multimodal, voice-enabled AI learning assistants handling student onboarding, live lecture transcription, and conversational tutoring using Whisper, Deepgram, and LiveKit. Backed by multi-threaded FastAPI + WebSocket services supporting concurrent real-time audio sessions in production classrooms.
Designed vector database architectures and retrieval-augmented generation (RAG) systems handling large-scale multimodal data and embeddings, deployed on AWS SageMaker and Bedrock via Terraform, cutting inference costs 40% while scaling to millions of monthly requests — all while keeping student data FERPA-compliant.