Mahendra User

I’m an Applied AI Engineer focused on shipping production-grade computer vision and LLM systems—especially multi-agent workflows and real-time inference. My goal is to turn cutting-edge research methods into reliable, measurable products with strong accuracy and performance.

In my current AI Engineering Intern role at Vishvena Techno Solutions, I architected an image-to-image pipeline using ControlNet, U-Net, Pix2Pix, CycleGANs, and Stable Diffusion, cutting inference time by 40% through TensorRT optimization and model quantization. I also built an end-to-end OCR solution integrating TROcr and Qwen Image I2L, achieving 95%+ accuracy across handwritten text, scanned PDFs, and multi-lingual documents (validated on 10,000+ samples), and reduced deployment memory footprint by 35% with mixed-precision training for sub-100ms edge inference.

I bring hands-on breadth from product-style development to AI research projects: I developed a multi-agent desktop automation system with 85% task success and improved reliability by reducing failure rate 60% using visual feedback and adaptive error recovery. I’ve also built RAG knowledge systems with 90%+ relevance accuracy, improved vector query latency from 800ms to 120ms, and integrated MCP for stateful, cross-platform AI interactions—backed by leadership organizing 200+ member AI/ML community initiatives and mentoring students from ideation to deployment.