Summary
Overview
Work History
Education
Skills
Projects
Awards
Patents And Publications
Timeline
Generic

Kartikeya Vats

Minneapolis

Summary

Accomplished AI scientist with expertise in deploying machine learning models at Target Inc, driving $60M in sales. Proficient in Python and deep learning, with a proven track record in innovative solutions. Strong communicator, adept at scaling data processes and enhancing model accuracy, saving $2.5M. Skilled in PyTorch and Vertex AI.

Overview

9
9
years of professional experience

Work History

Sr Data and AI Scientist, Product Availability

Target Inc
Minneapolis
01.2024 - Current
  • Created LightGBM models for predicting unknown out-of-stock items on the Target sales floor to trigger replenishment, leading to 60 million dollars in incremental sales.
  • Deployed ML models in production through Vertex AI
  • Set up distributing training through Modin and Ray clusters to scale the training data size from 10 million to 100 million records.
  • Fine-tuned RoBERTa on templatized natural language data, and trained it as an audit accuracy model, covering 10% more items with 20% more accuracy than the current third-party vendor's solution. The initiative will save Target $2.5 million and bring in additional sales.
  • Set up distributed inference for large language models on a Ray cluster for daily inference on Vertex AI, converting PyTorch models to Intel's OpenVINO framework for speedup via ONNX.

AI Science Intern, Search Relevance

Target Inc
Minneapolis
06.2023 - 09.2023
  • Optimized TextCNN-based classifiers for various product signals, such as Brand, Occasion, etc., to capture explicit query intent.
  • Implemented recency-based exponential decay for automatic product boosting to capture implicit query intent, to capture seasonality.
  • Fed the boosted results through explicit and implicit user query intent into the downstream learning to rank framework

Senior Data Scientist, Artificial Intelligence

IBM
Bangalore
08.2016 - 08.2022
  • Developed and deployed a recommendation system for a Korean fashion conglomerate.
  • Implemented Retrieval and Ranking modules for a novel multimodal deep learning-based recommendation system architecture (research paper, ACM CODS COMADS - dl.acm.org/doi/10.1145/3297001.3297035).
  • Implemented and patented a novel implementation of hybrid collaborative filtering.
  • Implemented and patented a pre-launch offline evaluation module for recommendation efficacy using PySpark. Evaluated the performance using MAP and NDCG.
  • Developed a Locality Sensitive Hashing algorithm for a 2.3M customer dataset, with incremental update capability, using PySpark.
  • Created a 'next best attribute' predictor leveraging Shannon's entropy to improve the user conversation experience and provide only relevant product space as input for the core recommendation engine.
  • Created a quantization-aware training pipeline for various deep learning models that were installed in the head unit of a Mercedes-Benz car.
  • Optimized YOLO, Faster-RCNN, and Mask-RCNN object detection models for Intel edge devices using Intel OpenVINO.
  • Employed OpenPose for social distancing checks in surveillance streams, deployed on the Jetson TX-2.
  • Created a distracted driver detection system for deployment in Raspberry Pi-based in-car systems using TensorFlow Lite.

Education

Masters of Science - Artificial Intelligence

Northwestern University
Evanston, Illinois
12.2023

Bachelors of Technology - Computer Science

NIIT University
Rajasthan, India
08.2016

Skills

  • Pytorch
  • Tensorflow/Keras
  • Transformers
  • Python
  • SQL
  • ONNX
  • Tensorflow-lite
  • OpenVINO
  • Vertex AI
  • Machine learning
  • Computer Vision
  • Natural language processing
  • Deep learning
  • Deep Learning Inference

Projects

  • ConvBERT - Method to capture hierarchical attention, Augmented self-attention with novel TextCNN inspired attention mechanism to capture hierarchical attention on top of regular self-attention.
  • Exploring the robustness of Large Language Model Watermarks, Replicated State of the Art paper results to watermark large language models generated text. Evaluated the robustness of watermarks in LLM by performing translation attacks on the generated text.

Awards

  • Best Paper Award, Demo Track in ACM CODS-COMADS - For work on recommendation systems, the paper was ranked as the best paper in demo track.
  • IBM Best Graduate hire award
  • Invention Plateau Award, For filing more than 4 patents.

Patents And Publications

  • Recommendence and Fashionsence: Online Fashion Advisor for Offline Experience, ACM-CODS COMAD, 2019, https://dl.acm.org/doi/10.1145/3297001.3297035
  • Learning roadmaps in unstructured text corpora based on topical basicness and advancedness measures, USA, 2022, US Patent 11,366,967
  • Deep Cognitive Constrained Filtering for Product Recommendation, USA, 2022, US Patent 11,532,025
  • Recommender System Evaluation using Time-Travelled Bagged SIM@K, USA, 2019, US Patent App. 16/830,926
  • UNSUPERVISED CONTEXTUAL LABEL PROPAGATION AND SCORING, USA, 2022, US Patent 11,526,707
  • Detecting Contiguous Defect Regions of a Physical Object from Captured Images of the Object, USA, 2025, US Patent App. 18/502,989
  • AI Driven Smart Patient Labeling System, USA, 2025, US Patent 12,293,845
  • Harmony: A Measure of Global Connectedness of Nodes in Signed and Unsigned Networks, USA, 2023, US Patent 11,849,342

Timeline

Sr Data and AI Scientist, Product Availability

Target Inc
01.2024 - Current

AI Science Intern, Search Relevance

Target Inc
06.2023 - 09.2023

Senior Data Scientist, Artificial Intelligence

IBM
08.2016 - 08.2022

Masters of Science - Artificial Intelligence

Northwestern University

Bachelors of Technology - Computer Science

NIIT University

Similar Profiles

CREATE PROFILE
Kartikeya Vats