Experienced in Machine Learning, NLP, and LLM-based systems, with a focus on building scalable, production-ready AI solutions. Worked on OCR and RAG pipelines at IIT Delhi. Proficient in Python, PyTorch, Hugging Face, OpenCV, and FastAPI.
Get to know me better
Experienced in Machine Learning, NLP, and LLM-based systems, with a focus on building scalable, production-ready AI solutions. Worked on OCR and Retrieval-Augmented Generation (RAG) pipelines at IIT Delhi for large-scale document understanding.
Proficient in Python, PyTorch, Hugging Face, OpenCV, and FastAPI, with strong experience in developing, optimizing, and deploying end-to-end AI systems using vector databases like FAISS and Pinecone. Currently pursuing BTech (Hons) in CSE - AI from CSVTU, Bhilai.
Outside of AI development, my hobby is playing badminton. I also run a YouTube channel where I upload vlogs and random study tutorials. Additionally, I manage an Instagram page @atrexplains where I post about various useful websites and tech insights.
Download ResumeCheck out my recent work
Built ML/DL models using MFCC, pitch, and energy features. Applied SVM, Random Forest, CNN, and wav2vec2 embeddings. Improved performance using augmentation and feature engineering. Accepted at RECAAP 2026 (IIT Palakkad).
Built OCR + ML pipeline for extracting structured data from scanned archaeological documents. Enabled semantic search using embeddings and hybrid retrieval with FAISS/Pinecone and Apache Solr. Co-authored SARCH paper with IIT Delhi.
Built LLM-based study assistant using LLaMA 3 and RAG architecture. Integrated FAISS/Pinecone for efficient retrieval. Enabled multimodal querying from PDFs, images, and YouTube content for comprehensive exam preparation.
Fine-tuned Mistral 7B using QLoRA for domain-specific healthcare NLP tasks. Integrated OCR pipeline for extracting insights from medical reports. Built AI chatbot that improved patient query resolution by 40%.
Built real-time audio recognition system using ACRCloud API. Designed end-to-end pipeline: audio capture → feature extraction → matching → metadata retrieval for instant song identification.
Modern, animated portfolio website featuring advanced CSS animations, smooth transitions, dark/light mode, and an immersive user experience.
Technologies I work with
Let's work together