Aug 2021 – Feb 2022 · 6 mos · San Francisco Bay Area

• Worked on building GUI elements (scroll-bar, tables) detection in the Electronic Heathcare records for Robotic process automation. This includes data collection and labelling, experimentation, tracking, deployment and monitoring. .
• Build a drift detection model to evaluate data drift in the production models, and trigger alert using Sentry.
• Tech stack : Fast API, Docker, Kubernetes, Postgresql, Pytorch (Detectron2), Sentry, GitHub, Codefresh, Streamlit, AWS (Lambda, Cloudwatch, EKS, ECS, Codeartifact, RDS etc.)

Mar 2018 – Aug 2021 · 3 yrs 5 mos · Mountain View, California

• AI Lead and founding Engineer (Google incubator startup)
• Built exercise recognition and rep counting model (over custom Posenet) and hand gesture recognition in live workout videos for live score and rank of users, and pitched this as a product differentiator.
• Lead the crowd-computing efforts to collect exercises dataset from Youtube, extract, filter and convert it into trainable dataset. This dataset is shared across research team at Google and is used to power for other similar vision models like Movenet.
• Lead a team of five 20%ers, in the area of analysis, data pipeline, ML research to build sophisticated 3D models (over Posenet) which can do pixel level depth detection (trained on 3D-Simulated dataset).
• Built ML pipeline from Youtube video extractions, data annotation, filtering, pre-processing, training, model optimization, real user and device testing for video understanding.
• Did tf model optimization, integration in Angular web app and tuning to fit the user behaviour and UX.
• Developed the data pipeline and provided useful insights using GA, Firebase, BigQuery, SQL etc. to get insights about the user behaviour, video stats and led the efforts for analytics and insights for product improvement.
• Firebase Predictions (March'18- Nov'19)
• Role: ML Engineer-III
• Worked on ML pipeline. feature engineering, modeling and backend aspect to improve the customer segmentation for better churn and spend projections for app developers.
• Mentored one Phd ML intern to explore RL for increasing developers conversion of buying tokens(spending) and reducing churn. This is converted into a fully functional product later.
• Talks links:
• https://www.meetup.com/Hacking-Machine-Learning/events/264087200/
• https://www.meetup.com/fr-FR/GDG-Asturias/events/263783888/ (GDG Asturias, Spain 2019)
• https://www.meetup.com/gdg-singapore/events/256624684/ (GDG Singapore 2018 )
• https://gdg.community.dev/events/details/google-gdg-asturias-presents-firebase-asturias-day/

Jan 2017 – Mar 2018 · 1 yr 2 mos · Greater Seattle Area

• Part of Data science and Machine Learning team at Zillow.
• Working on scalable and optimization of machine learning pipeline that runs Zestimate. Worked on the pre-processing and data cleaning part in Apache Spark on EMR.
• Designed and wrote a deep learning model in tensor flow to find similar houses from the Zillow image database, detect and link the house objects from external e-commerce website. Used external library for object detection. Designed and implemented image similarity engine by training an CNN and using it as encoder (using Fully connected layer as image encoder)
• Worked on developing the complete pipeline(from data analysis, feature extraction, transformation, engineering and ranking etc.) to classify Mobile API request as legit or not with 0.008% rate of false positive. Used Spark Mlib for modeling, cross validation and spark SQL+Pandas+matplotlib for pre-processing, analysis and feature engineering part.

May 2016 – Aug 2016 · 3 mos · Greater Seattle Area

• Built an anomaly detection model for online time series data using Bayesian online changepoint detection as underlying technique to divide consecutive data points into same sequences.
• Used K mean clustering for defining the number of states/distribution present and identify outliers/anomaly, and used Hidden Markov model to learn pattern/state sequences.
• Implemented this algorithm on streaming data using Apache Kafka, Flink, Cassandra and HDFS.
• Also learned about the implementation of 6 layered RNN to predict the future data using tensorflow.

Feb 2016 – May 2016 · 3 mos · Greater Pittsburgh Area

• Worked on the development of advanced Machine Learning model like GcFlasso, GwLasso, Linear Mixed Models to do the SNPs and phenotype prediction. Developed algorithms like grid search,brent's search and Smoothing Proximal Gradient descents to find the optimal parameters by minimizing the cost functions of the above mentioned models.
• Work done in C++ using Eigen Library.
• Project : GenAMap
• Link : http://www.sailing.cs.cmu.edu/main/?page_id=275

Nov 2012 – Jul 2015 · 2 yrs 8 mos

• General role and Responsibilities (WCDMA Physical Layer – Air Interface)
• Solving SW bugs related to physical layer, customer issues, field test logs analysis, KPI’s, developing customer specific feature, optimization of existing SW algorithms for better performance, UMTS/HSPA+ protocols related issues, Mobility management for multi carrier etc.
• Software design (Physical Layer), development, testing, integration of Rel-9/10 features i.e DC-HSUPA and DB-DC. Inter-operability testing with network vendor i.e Huawei, Shanghai and Ericsson.
• Proposed and implemented many WCDMA Physical layer optimization to Customers like Apple/HTC to boost their performance related to Intra-frequency and Inter-frequency measurement which further results in better soft/hard handover, reduction in call drop/RL Failure.
• Understand and fixes issues related to SW-FW interface, race condition, RTOS resource starvation and deadlock, multi-tasking and thread level parallelism on Qualcomm modem.
• Represented Qualcomm in Rel-9 DCHSUPA IODT and IOT with Huawei, Shanghai.
• Syncrod (Startup project funded by Qualcomm innovation program)
• Worked in a team of 7 members in Qualcomm, San Diego to build an ecosystem to automate user data backup by storing the encrypted data among peers and thus eliminating monthly payment and third party storage.
• During prototyping phase, I worked on the UI design, android mobile apps development(Eclipse) and algorithms development for data encryption and retrieval (C++).
• Showcased the working demo and business slides to Qualcomm’s executive Board at San Diego HQ in July’13.

Jun 2011 – Oct 2012 · 1 yr 4 mos

• Gained In-depth understanding of WCDMA Physical Layer Architecture/Software implementation and its interaction with higher layers.
• Analyzed and fixed many customer(Apple/Samsung/HTC etc.) issues related to cell camping, Acquisition failure, call drops, energy Issues, Measurement event reporting, handovers, specs related issues, performance optimization etc. along with generic OS issues and Software-Firmware Interaction.

May 2010 – Jul 2010 · 2 mos · Paris (Gif-Sur-Yvette), France

• Optimization of GPU algorithm by reducing the time taken to transfer data between CPU and GPU. This is used in application of 3-D Computed Tomography.
• Report: https://l2s.centralesupelec.fr/wp-content/uploads/gac-nicolas/Rapport_stage_Aditya_GAUTAM_2010.pdf