Kunal Kumar Email & Phone Number

Santa Clara, California, US

Santa Clara, California, US

• As a part of Machine Learning Solutions Team in Artificial Intelligence Group at AMD, my role in the internship were:
• Designed a dockerized pipeline to quantify the efficiency of Post-Training Quantized (PTQ) NLP models, resulting in a 41% & 75% reduction in model size and a 2x & 1.77x speedup in inference for PyTorch & ONNX models.
• Fine-tuned LLMs like BERT & Llama on SQuAD and GLUE dataset achieving 85% accuracy for quantization task.
• Exported trained weights from Deep Learning Recommendation Model (DLRM) model to ONNX & optimized the model via quantization using AI compiler.
• Transformed the model into ONNX format, performed operator analysis, and executed model inference.Exposure: Natural Language Processing(NLP), Computer Vision, Deep Learning, Recommender System, Docker, Linux

Austin, Texas, US

• Title: Operating Software with Grounded Language Agents.Industry: Machine Learning Research Group, Oracle LabsIndustry Mentor: Ari KobrenPhD Mentor: Dhruvesh PatelSupervisor: Prof. Andrew McCallumDetails:
• Developed a Grounded Language Learning model using FLAN-T5 LLM in Pytorch for instruction-based finetuning onOracle dataset with a 15% increase in accuracy metrics, including Token F1 and Token Exact Match.
• Incorporated various prompting techniques, including Question-Answer, Machine Translation, & Chain-of-Thought toenhance model evaluation with few-shot learning, improving overall model performance by 20%.
• Automated dataset creation using web scrapping tools on Oracle documentation streamlining it by 85%.Exposure: Python, NLP, LLM, Pytorch, Transformers, TensorFlow, Embeddings, Web Scrapping

New York, NY, US

• Orchestrated successful project deliveries within CI/CD practice, yielding in a 20% enhancement in project delivery efficiency and a 10% surge in production stability collaborating with the Risk & Forecasting Solutions.
• Automated the generation of Value at Risk (VaR) and Stress numbers using Python, improving data calculation accuracy and reducing manual effort by 70%.
• Optimized code deployment workflow through parallel processing strategies by 25% & improving overall performance.
• Generated synthetic data using Random Forest model, significantly boosting the precision of historical VaR calculations.
• Implemented a time series model for VaR calculation & backtesting, accurately forecasting volatility & estimating VaR under extreme conditions.

New York, NY, US

Allahabad, Uttar Pradesh, IN

Summer Research Internship in SILP(Signal Processing, Image Processing, and Language Processing) Lab.Project: Development of a syllable based unit selection method for Hindi text to speech generation. (A part of Dialogue System)About: This was 2 months of research internship under SILP Lab. The key area to focus was Natural Language Processing, Machine.

Master Of Science - Ms (Conc. In Data Science), Computer Science

Integrated Dual Degree (Idd), Electrical And Electronics Engineering

10+2 Level, Cbse ( Grade - 92.8% )

10Th Standard, Icse ( Grade - 92% )