Rajat Raj

I’m an MCA student at RV College of Engineering and an AI/ML Intern who builds end-to-end machine learning projects. I enjoy moving from data preprocessing and model training to working, user-facing outputs.

In my internship at Centre of Competence in Visual Computing (CoCVC), I built a video-to-text summarisation pipeline. I extracted audio with FFmpeg, transcribed with OpenAI Whisper using chunking and VAD, then applied LoRA fine-tuning via HuggingFace PEFT on TinyLlama-1.1B-Chat with BitsAndBytes 8-bit quantisation for CPU constraints.

In projects, I combined deep learning with practical interfaces and evaluation. For medical image segmentation, I implemented a CNN organ segmentation model trained on CHAOS MRI and refined boundaries using A* pathfinding with CNN-predicted cost maps, achieving a Liver Dice score of 0.87 and adding a Gradio inference interface on Kaggle.

I also built a multilingual text-to-image generation pipeline using Meta NLLB-200 for translation, Stable Diffusion XL for image generation, and GFPGAN for face enhancement, wrapped in Gradio with langdetect. For healthcare, I trained a Random Forest model with GridSearchCV, performed EDA and feature engineering (including Length-of-Stay), and created a Tkinter GUI for real-time predictions.