Spotify | Deep Learning for 173M user-trend analysis
Led data handling, analysis, deep learning design and productionization
of seq-to-seq model for time-series analysis of 173 million Spotify users.
Achieved 10-20% Qini AUC, improving 50% over existing 6-7% XGBoost
production model.
Led the end-to-end lifecycle, from data creation in BigQuery, data
preprocessing and analysis, model design in TensorFlow / PyTorch,
training on Vertex AI, test using PR-AUCs and Qini curve AUCs,
model deployment to production pipelines in Kubeflow.
Deployed models to production pipelines with Kubeflow, Docker and Google Cloud Platform.
Jun 2022 - Aug 2022
New York, NY
Machine Learning Engineer Intern
Panoskin
| Computer Vision Research and Development
Led the development end-to-end XGBoost and RandomForest pipelines with
91% precision, accuracy: for creating reliable 3D Google Map tours from
360 images, IMUs, Google Map APIs.
Modelled the pipeline end-to-end to extract data from computer vision data,
create features from the data, search best features, train and test models
for confidence predictions on 3D image data.
Working on machine learning for improving 360 image based google map tours.
Apr 2022 - Jun 2022
Chicago, IL (Remote)
Machine Learning Research Assistant
Computer Science Department, Carnegie Mellon University
| Deep learning research
Published a conference paper with the EAAI symposium as part of the
AAAI 2022,
The paper can be found
here.
Developing word-embedding visualization software for explainable natural language processing.
The software is deployed at a national level for setting AI guidelines and teaching AI in schools in Georgia.
Sep 2021 - Present
Pittsburgh, PA
Machine Learning Instructor
Inspirit AI
| AI innovators - graduate program mentorship
Led a team of 10 undergraduate, graduate students to build a music recommender system project on
Spotify data APIs using content filtering, applying BERT embeddings to audio datasets and music data.
Led a series of machine learning lectures from basics to deep learning applications in vision and language
models using appropriate neural network architectures.
Conducted a professional development session for networking and working with the machine learning industry.
Dec 2021
Palo Alto, CA (Remote)
Deep Learning Research Fellow
College of Engineering, Carnegie Mellon University
| Deep Learning Research
Designed a TensorFlow based neural network solver for solving flow problems with computational efficiency.
Optimized the physics informed neural network to generate accurate predictions using tailored loss functions using gradients extracted from backpropagation and autograd functionalities of TensorFlow.
Jun 2021 - Aug 2021
Pittsburgh, PA
Computer Vision Research Student
CERLAB, Carnegie Mellon University
| Computer vision research
Used convolutional neural networks in conjunction with geometrical algorithms for 3D pose estimation in UAV quadrotor.
Performed 3D bounding box detection using YOLOv3 for navigation of the quadrotor.
Jan 2021 - Jun 2021
Pittsburgh, PA
Machine Learning Instructor
Camp K12 Inc.
| Machine Learning program mentorship
Taught around 100 students machine learning basics using JavaScript
Mentored image, video, sound detection projects, chatbots and 3D interactive apps.
Projects were created using JavaScript, Google Teachable Machine, ml5, p5 libraries, proprietary and open source JS based platforms
Sep 2018 - Jan 2019
Mumbai, India
Design & Controls Engineer
PMV Electric Pvt. Ltd.
| Electric vehicle engineering
Modelled the cruise control system for a battery electric vehicle by programming a PID controller
Initially simulated a dynamic vehicle model with basic manual vehicle controls.
The model was deployed on a Carla server and tested on a simulated vehicle.
Finally the programmed controller logic was deployed and tested on a C++ based Curtis EV controller.
Also designed and optimized body frame of the vehicle using ANSYS and SolidWorks based crash simulations to get a 51.38% weight reduction
Vehicle completed 600 km of successful testing based on these designs. You can see the launched product here.
Jun 2018 - Sep 2020
Mumbai, India
Project Department Intern
Mahindra and Mahindra Limited
| new launches
Worked on an upcoming vehicle prototype, now launched as the Mahindra Marazzo
Created solutions to 107 different fouling problems using electro-mechanical engineering techniques.