Harshveer Singh

Enhanced production model metrics by approximately 11% for US-based NBFC customers using a custom RoBERTa-based multilingual model. Developed and integrated CI/CD pipelines for seamless model deployment, reducing manual intervention and accelerating update cycles.

An in-house multi modal large model (finetuned Llama 9B quantized) with a RAG layer to provide candlestick
chart analysis in various formats.
 Production stack nodejs, CSS/JS, weaviate vector db, pytorch, llammacpp, LoRA, PEFT
 Have a healthy user base with rich community feedback.
 Also launched on Product Hunt [https://www.producthunt.com/products/chartchatai]

Improved inter-sentence and intra-sentence cohesion measuring pipeline precision by 25% using PyTorch, Huggingface, and NumPy. Implemented a production-ready email formality checking pipeline using improvised LDA for topic modeling.

Worked on improving Natural Language Inference (NLI) models, enhancing cohesion detection accuracy on English text by approximately 16%. Gained practical experience in applying NLP techniques to real-world text analysis problems.

Worked on building a quasi-linear attention model to isolate 'interesting' events from background noise during proton-Low-Z target collisions. Contributed to research in deep learning applications for high-energy physics.

Provided an analytical study on the success of Batch Normalization, contributing to the understanding of deep learning optimization techniques. Engaged in research within the Department of Mathematics.

Activities and societies: Research Intern, Department of Mathematics, Research Intern, Deep Learning Research Intern

Studied core concepts in computer science and engineering. Engaged in research activities and projects related to AI and machine learning.