MS in Statistics

Aug 2019 - Dec 2020

University of Illinois, Urbana-Champaign

GPA: 4.0

Statistical Modeling II, Statistical Learning, Time Series Analysis, Statistical Data Management

Bachelor of Technology, Computer Science

2012 - 2016

Jamia Millia Islamia, New Delhi

GPA: 9.6

Artificial Intelligence, Data Structures, Object Oriented Programming, Database Management Systems, Operating Systems, Advance Computer Networks.

Data Science Manager AI Foundations

July ’24 - Present

Capital One

• AI Safety – Led the development and fine-tuning of Large Language Models (LLMs) using Chain-of-Thought (CoT) reasoning techniques to establish robust guardrails for both input and output. This initiative safeguarded company-wide chatbot systems from producing or responding to inappropriate, harmful, or malicious content, enhancing the ethical standards and user safety in AI-driven interactions.
• Red-Teaming – Conducted comprehensive red-teaming exercises on LLMs, employing advanced techniques such as Tree of Attacks with Pruning, Crescendo, and Encoding attacks to test and expose vulnerabilities systematically. These rigorous evaluations helped identify potential weaknesses in the models' responses, contributing to improved security measures and the overall resilience of AI systems.

Data Scientist Enterprise Intelligence

Jan ’23 - July '24

Home Depot

• Creating conversational AI solutions for Interactive Voice Response (IVR) and Messaging systems within Home Depot. This involves intent classification, slot fitting, and smart reply predictions.
• Submitted a paper to EMNLP 2024 titled ”Enhancing Task-Oriented Dialogue Systems: Semantic Aware Nearest Neighbors for Real-time Intent Classification” which aims to detect customers intent in a task-oriented dialogue system.

Applied Scientist Alexa AI

Aug ’21 - Jan ’23

Amazon

• Developed multi-modal SoTA cascaded Neural Turing Machines (NTM) based network for multi-turn fashion image retrieval with an overall improvement of 50% over previous benchmarks.
  • Using multi-turn transactions of an image and feedback text, the model retrieves the best matched image based on the interactions between the user and machine.
  This was deployed as a chatbot feature in Alexa Fashion domain and is in beta testing. Paper accepted in ICCV 2023.
• Worked on gesture recognition for Augmented Reality (AR) on hands using object detector, landmarks extractor, and Kalman filter tracking using mobilenetV2 and efficientnet resulting in gains of over 5%.

Data Scientist

Feb ’21 - Aug ’21

Ciitizen Corporation

• Built framework for Named Entity Recognition (NER) of seminal entities such as Diagnostic and Therapeutic Procedures using LSTM from clinical pdfs.
Developed Tensorflow and Kubeflow based CNN model to segregate clinical texts on the basis of importance, thereby reducing the manual read time by 2X hours.

Machine Learning Engineer Financial Modeling Group

Feb ’18 - Aug ’19

Blackrock

• Built end-to-end word-embedding model similar to word2vec . Enabled large scale training using Ignite for async updates across multiple nodes. Deployed the code using Docker image and Kubernetes
• Developed a novel word-level Entity Linking/Disambiguation model by using BERT and bi-LSTMs surpassing SoTA on AIDA test dataset by 2%. Model deployed using tensorlfow-serving. Accepted in AACL.

Data Scientist

June ’16 - Sept ’17

Scry Analytics

• Relation Extraction: Developed a Convolution Neural Network(CNN) based deep learning model to identify relationships between entities in a text. Achieved F1 score of 85% surpassing previous feature-based models by 15%.
• Implemented Big Data pipelines in Spark for fast retrieval and processing of data residing in HBase. Reduced the pipeline execution time from 3 days to 6 hours by switching the entire pipeline from Map-Reduce to Spark.

NLP Research Intern

Aug ’19 - Dec ’20

Inference Analytics

• Created an end to end model to perform aspect-based sentiment analysis using pre-trained encoders and topic modeling techniques (LDA, Rake).

Machine Learning Engineer Intern HAMR Write Design Team

May ’20 - Aug ’20

Inference Analytics

• Designed a web-based Machine Learning pipeline for data segmentation, wrangling and ranking.