I built and scaled a GPT-style transformer training pipeline from ~1K to 3B+ parameters across an NVIDIA Jetson Orin Nano and Google Colab A100 GPUs, and published model checkpoints to Hugging Face.
For YODA, I designed a multi-agent marketing platform with FastAPI and PostgreSQL, integrating pgvector and a Next.js client portal. It supported a JEE Mains 2026 lead-generation campaign with 500 leads and a ₹50K budget.
At Avintix Systems Private Limited, I manage patching across 150+ Windows and Linux EC2 instances in three AWS environments, maintaining 98%+ patch compliance. Previously at Arcnas Software Solutions, I automated CloudFormation provisioning across AWS accounts and implemented SSM Session Manager to remove inbound SSH/RDP exposure.
