May 2023 – Present · 2 yrs 10 mos

• My passion of creating Travel content and sharing itineraries on Instagram with many reels of over 1Million views

Aug 2022 – Feb 2023 · 6 mos · San Francisco Bay Area

• Key contributor for Omnimizer PyTorch library for hardware-efficient Neural Architecture Search (NAS) and pruning
• Improved the torch.fx based tracer to support new PyTorch modules and operators for NAS
• Built tools for understanding and visualizing module/operator-level latencies of a PyTorch / ONNX model for NVIDIA TensorRT and Qualcomm SNPE edge devices
• Contributed to the MVP of a job management dashboard to track experiments, visualize ONNX models, and latency insights
• Lead the technical product demos at executive summits and AI conferences

Feb 2021 – Jul 2022 · 1 yr 5 mos · New York, United States

• 1. Self-service ML Model Remediation and Maintenance Pipeline:
• Built a generalized platform allowing AI teams to setup their self-service model maintenance workflows consisting of automated annotation pipeline setup, model training, evaluation, and deployment using Apache Airflow
• Significantly reduced KTLO time in remediating issues in deployed models by empowering domain experts to perform model maintenance with minimal engineering support
• 2. Topic Tagging in Financial Documents
• Enriched documents like analyst research reports with trends and topics per paragraph using TextCNN and BERT models
• Setup data pipelines for sampling and cleaning paragraphs using PySpark for collecting annotations
• 3. Government Contract Recommender system:
• Setup model training infrastructure on AWS SageMaker using data from Athena for recommending government contracts to Bloomberg Government’s (BGOV) client organizations
• Created Docker images to deploy trained models to production using a FastApi server

May 2020 – Aug 2020 · 3 mos · Greater Seattle Area

• Deployed Java APIs to AWS cloud for providing a preview of tasks to be performed by technicians in Amazon data centers
• Parsed complex BPMN Workflows extracted from Amazon Dynamo DB to identify expected order of task execution
• Tested code with Unit, Integration and Load testing with JUnit, Mockito, and TestNG

Aug 2019 – Dec 2020 · 1 yr 4 mos · Urbana-Champaign Area

• Experimented a Visual Attention-based Model in PyTorch for novel Webpage Object Detection formulation
• Utilized contextual information using visual features of ordered web elements extracted using Resnet18
• Created largest public labeled dataset of 7.7k product webpage screenshots
• Achieved 95% accuracy for product Price detection (8.5% above Fast R-CNN) and interpreted Attention Visualizations Attention Visualizations

Jan 2019 – Jul 2019 · 6 mos · Bangalore

• Implemented Graph Recurrent Neural Networks from scratch in TensorFlow to Learn Embeddings for 300,000 entities in a heterogeneous graph
• Collaborated with a team of 10 to design a novel Deep Neural Net architecture for recommending messages in MSTeams
• Outperformed Matrix Factorization Methods like SVD & SVD++ on 2 benchmark rating prediction datasets
• Achieved comparable or better results than Graph Convolutional Networks (GCNs) on 8 benchmark datasets for node classification, ranking, and rating prediction tasks

May 2018 – Jul 2018 · 2 mos · Hyderabad, Telangana, India

• Worked on Budget Enhancements for Expense Management System by adding back-end services in Java to exclude centers for Budget Allocation Process and comparing budgeted inputs with actual expenses by extrapolating data.
• Designed database schema for efficiently storing center exclusion rules with dynamic SQL using My Batis, and wrote stored procedures in MS SQL Server for modifying database contents.
• Made a UI using JavaScript for creating/modifying center exclusions, uploading excel inputs, and showing comparison grid.
• Wrote about 100 unit test cases in JUnit and Mockito and increased test coverage by 5%
• Learned Continuous Integration using Jenkins and Software Development Life Cycle process.
• Finished roadmap project before the expected date of completion.

Jan 2018 – Dec 2018 · 11 mos · Greater Hyderabad Area

• Analyzed Twitter Stream for Event Detection by leveraging Wikipedia
• Segmented tweets and hash-tags; applied Jarvis-Patrick clustering; summarized newsworthy events in Python
• Achieved a precision of 88.12% (absolute improvement of 8%)

Aug 2017 – May 2018 · 9 mos · Greater Hyderabad Area

• Organized ’Dandiya Night’ - a cultural dance event at BITS Pilani, Hyderabad which was attended by over 1000 students for which a fund of 30,000 INR was raised
• Arranged food stalls of Gujarati Delicacies
• Conducted several traditional dance workshops

May 2017 – Jul 2017 · 2 mos · Worli, Mumbai, India

• Worked on the front end of a location tracking device using Google Maps JavaScript API
• Made a blog website using HTML, CSS, JavaScript, and PHP