Mitesh J

I'm an AI Engineer focused on building reliable AI systems that perform under real-world constraints, with experience spanning Computer Vision, Agentic AI, RAG systems, and LLM evaluation. My work centers on improving accuracy, reducing latency, minimizing hallucinations, and establishing robust evaluation frameworks for production AI applications.

At Emergent, I worked as an Agent Reliability Engineer, designing evaluation frameworks to assess planning accuracy, reasoning quality, tool utilization, and prompt adherence for AI agents. By refining prompts, model configurations, and evaluation pipelines, I improved task completion rates by 30% while reducing hallucinations in production systems. I also automated regression, integration, and unit testing using Playwright and PyTest, reducing manual QA effort by 40%.

At Centurion Robotics, I developed real-time computer vision pipelines for drone-based surveillance and emergency response. I engineered an edge inference system integrating YOLOv8, DeepSORT, Kalman filtering, and GPS mapping to improve multi-target tracking and survivor prioritization. Additionally, I implemented pose-based feature extraction and machine learning classification models that achieved approximately 85% survival-state detection accuracy, while optimizing processing pipelines to reduce latency by 35%.

My project experience includes AI-powered legal contract analysis using transformer-based NLP models, multimodal threat detection systems with real-time alerting, RAG-based AI applications, and large-scale agent evaluation platforms. I am passionate about building AI systems that are not only intelligent but also measurable, dependable, and production-ready.