Kunal Banerjee

About Kunal Banerjee

My key strength is Research-to-Realization mindset:> Explore SOTA research> Integrate the potential researches into new/existing tools for PoC> Showcase the findings to leadership> Convert successful PoCs into industrial offeringsI have recently joined Solventum as the founding member of the AI Science team in India, where we focus on applying AI - particularly Generative AI - to solve complex problems in the healthcare domain for global markets, with a strong emphasis on the EMEA region.Prior to this, I was a Principal Data Scientist at Walmart, working on a wide range of initiatives including text extraction from images, anomaly prediction in production channels, and the introduction of sequential testing within the A/B experimentation framework. I also led the development of the world’s first super-resolution technique optimized for improving text legibility in low-resolution images. Until my departure, I was responsible for the implementation and monitoring of company-wide Responsible GenAI. Through this role, I gained extensive experience across Computer Vision, Natural Language Processing, Recommender Systems, A/B Testing, AutoML, and AI Governance.Earlier, I was a Research Scientist at Intel Labs, where my research focused on kernel optimization of deep learning workloads on Intel architectures. My work on optimized implementations of convolution (Winograd), RNNs, LSTMs, and GRUs is available in open-source libraries such as, LIBXSMM and Intel MKL-DNN. These libraries have been adopted across major frameworks and products including PyTorch, TensorFlow, Caffe, MS CNTK, Apache MXNet, Chainer, and OpenVINO, enabling significant performance improvements on Intel silicon.My research interests also include low-precision deep neural networks. In collaboration with colleagues at Intel Labs, I co-developed Ternary Residual Networks, which use 8-bit activations and 2-bit weights (with residual edges where required). I am also a co-author of the foundational paper on the BFLOAT16 datatype, a collaborative effort between Intel and Facebook. Additionally, I contributed to Intel’s deep learning training accelerator as part of the Intel Artificial Intelligence Products Group.Before joining Intel, I earned my PhD in Computer Science from IIT Kharagpur. My doctoral research was supported by multiple fellowships, and my dissertation received the Best PhD Thesis Award from several venues, including VLSI Design, ISVLSI, and the India Electronics & Semiconductor Association (IESA).Please see my personal website for more information.

Experience

1. Principal Ai Scientist

   Solventum

   Dec 2025 — Present · Bengaluru, IN

Education

• Heritage Institute of Technology, Kolkata

  Bachelor of Technology (B.Tech.), Computer Science and Engineering

  2004 — 2008

• Ramakrishna Mission Residential College

  Higher Secondary

  2002 — 2004

• Lycee

  ICSE

  2022

• Indian Institute of Technology, Kharagpur

  Doctor of Philosophy (PhD), Computer Science

  2010 — 2015

Skills

• Oracle Pl/Sql Development
• Algorithms
• Javascript
• Fpga
• Formal Verification
• Visual Basic
• C
• Programming
• Verilog
• Html Scripting
• C++
• Python
• Latex
• Linux
• Embedded Systems
• Data Structures
• Machine Learning
• Java
• Matlab
• Simulations
• Formal Systems
• Computer Science