Maulik Chhetri

Designed and deployed a novel 90%-accurate agentic-RAG QA framework for Simulink models using FastAPI, React, and Docker. Benchmark-tested 6+ LLM architectures using RAGAS and LLMEval, incorporating iterative feedback from 10+ Infineon engineers. Collaborated with cross-functional domain experts to structure model data and led training sessions to boost engineering productivity.

Developed a data augmentation pipeline to cut costly RADAR camera recordings while enhancing a 3D CNN presence detection model to over 90% accuracy across all scenarios. Implemented Grad-CAM explainability and built a reusable evaluation framework benchmarking neural networks across 4 latent-space metrics. Researched adversarial robustness for the project and presented key findings to 50+ engineers at an internal AI conference.

1. Researched and implemented advanced deep neural networks for instance segmentation, semantic segmentation, and object detection using Python and TensorFlow. 2. Investigated Bayesian active learning techniques to enhance data efficiency. 3. Authored a comprehensive knowledge base for ML engineers, documenting models and optimization strategies.

1. Conducted bi-weekly tutorial on topics related to Linear Algebra, ML algorithms, Computer Vision and Robotics for 100+ students. 2. Collaborated with other TAs to design assignments and grade examinations.

1. Conducted weekly tutorials for 100+ students on basic to advanced concepts of Computer Science. 2. Collaborated with professor to grade examinations and assignments of all BSc. Computer Science Students.