Aarathi Vijayachandran

Intern Ai Ml Research @BMW Group

Munich, DE

Virtual Surface Analysis using Vision Transformers (ViT):• Developed an end-to-end AI/ML pipeline to detect and classify automotive surface defect severity.• Implemented 2 stratified dataset splitting strategies (video-level and image-level) on a Finite Element Method (FEM) simulation video dataset with 4 defect severity classes.• Fine-tuned and evaluated 6 vision transformer models (ViT, ViT-Hybrid, Swin, DeiT, BEiT, ConvNeXt) on both dataset splits, achieving over 90% accuracy and stable training on the image-level split.RAG Pipeline for Multilingual Defect Query Resolution:• Designed an end-to-end RAG pipeline to perform semantic search over Failure Tracking Management (FTM) tickets and generate solutions for users\' multilingual defect queries.• Processed FTM ticket XML data into an SQLite database, generated multilingual embedding textswith GPT-4.1, and encoded them into 768-dimensional vectors using MPNet Sentence Transformer.• Indexed embeddings in FAISS, applied Cosine Similarity-based semantic search to retrieve Top-10 most relevant FTM tickets per user query, and used GPT-4.1 to generate precise defect solutions.

Birla Institute of Technology and Science, Pilani

Master of Technology - MTech, Data Science and Engineering

College of Engineering Trivandrum

Bachelor of Technology - BTech, Electrical and Electronics Engineering

Otto-von-Guericke University Magdeburg

Master of Science - MS, Digital Engineering (Focus: AI & Data Science)