+1-236-***-****

adcp7s@r.postjobfree.com

Vancouver, BC

Abhi Savaliya

Data Scientist Machine Learning Engineer

https://www.linkedin.com/in/abhisavaliya/

https://github.com/abhisavaliya

Summary

• Data scientist and machine learning engineer with 1+ hands-on experience by executing data-driven solutions to increase scalability, performance of models & provide business insights. Passion for AI.

• Takes pride & excel in building innovative, creative solutions using AI & applied data science techniques. Experienced at predictive modeling, analyzing the data & knows the mathematics behind the techniques. Professional/Industry Experience

Data Scientist – Praxis Spinal Cord Institute (Part-Time) Jan 2020 – Present

• Responsible to implement Proof of Concepts (POCs) from statistical/mathematical research methods and research papers published at Springer, Journal of Applied Statistics.

• Performed multiple quantitative statistical data analysis by using p-values, normal distribution and PCA to perform feature selection.

• Prototype a model using TensorFlow to translate the linear Poisson regression into Non-Linear Poisson regression using neural networks. Results improved by 5-7%.

• Model predicts the number of risk factors of patient for the spinal cord injury using 1200 patient’s data.

• Data is count data which uses Linear Poisson Regression, Poisson & negative binomial distribution.

• Compared the statistical methods - Linear Poisson regression and Non-Linear Poisson regression to verify the effectiveness on the count data.

• Technology: TensorFlow, Python, R, Scikit-learn, pandas and matplotlib for data visualizations. Machine Learning Engineer – Ericsson Canada (Co-op) May 2019 – Dec 2019

• Driven the team of 4 strategically to the end goal by following agile development, communicate clearly & conducting sprints with business stakeholders, product managers and other teams.

• Focus on reducing the 5G network latency and improve business resources usage by supervised machine learning techniques and experiment using A/B testing on multiple configuration of data sets.

• Collaborate with multiple team members to understand the pipeline of predictive model.

• Prioritize the model to scalability and analytical implementation of model.

• Built entire architecture from the large-scale data collection to finalizing model including pre-processing & feature engineering. Dataset comprises over 4 million rows and 200+ features.

• Applied GBM, XGBoost & tools like AutoML (h2o) to build the classification model.

• Developed Ericsson's legacy algorithms to verify the simulated data.

• Statistical methods used–chi2, null hypothesis test (p-value) & PCA for feature selection.

• Exhibited the project in Ericsson Developer Conference 2019 in Ericsson Canada (Ottawa).

• Reduced latency by 38% by prototyping models, improve business impact and efficiency of the core 5G network. Reduced resources usage by 32% by model experiments.

• Apart from project, gained intense knowledge of Java OOP (Object Oriented Programming) & python scripting language concepts using "Head First Design Patterns" for software development – Ericsson Initiative

• Member of Toastmasters to improve communication skills and interpersonal skills.

• Technologies/Libraries: Python, spark, TensorFlow, Scikit-Learn, AutoML - h2o, SQL, statsmodels, Linux Education

Master’s in Computer Science, Big Data Specialization – Simon Fraser University Sept 2018 – Apr 2020

• Courses: Machine Learning, Big Data, Statistics, Algorithms

• GPA: 3.62

Bachelor’s in Computer Engineering – RAIT (University of Mumbai) Aug 2014 – Jun 2018

• Courses: Artificial Intelligence, Data mining and Analytics, Mathematics, Soft Computing, OOP

• CGPA: 8.71

Programming Skills

Programming Languages Python, Java, MySQL

Frameworks, Databases TensorFlow, AutoML, Apache Spark, Cassandra, REST API, Tableau, scipy, pandas Technical Skills Deep Learning, Statistical modeling, Data Science, Statistics, Predictive Modeling Research Experience

Reinforcement Learning with Transfer Learning on Atari 2600 Published – Feb 2019

• Investigated the feasibility of applying Deep Learning (CNN) and Transfer Learning in Atari games 2600 environments to gain efficient performance and reducing the training cost using Keras-RL

• Improved RL Agents’ (DQN, SARSA, DDQN) performance (77% by Reward) with hyperparameter tuning

• Applied various policies and optimizers to improve the gain rewards on the video games

• Link: https://www.ijariit.com/manuscript/improving-generalization-in-reinforcement-learning-on-atari- 2600-games/

Social Champion Identification with Machine Learning Published – Apr 2018

• NGOs are facing problems to look for individuals to promote & spread awareness. We help NGOs by deploying machine learning model for collaboration with potential, motivated donors & volunteers to help their social cause.

• Model uses k-means++ to divide users among 3 clusters based on Relevance, Reach & Resonance of tweets.

• Link: https://www.ijariit.com/manuscript/social-champion-identification/ Personal Development/Projects

Artificial Intelligence (AI) controller based on Computer Vision (LinkedIn Trending)

• Implemented state-of-the-art computer vision techniques to rebuild controller using gloves or just paper

• Gloves/paper controls the game using Image Processing which aims at improving user’s health and fitness

• Incorporated Google Cloud Platform (GCP) API for speech recognition to perform activities in games

• Used OpenCV library to detect gloves/paper ands provide analyzes in real-time using numpy operations

• Consulted 3 Physiotherapists in Ottawa to check the feasibility of the model and make it reliable

• Highly applauded by LinkedIn community. Trending on LinkedIn with 40000+ views and 1600+ likes

• Video Games: Need For Speed (Electronic Arts), Call of Duty

• Technologies: Computer Vision, OpenCV, Numpy, GCP Speech Recognition, Python, Git HSV Calibration Open Source Library

• Developed custom library to contribute to open source and computer vision community

• A lite library for easy background subtraction and gets accurate HSV (RGB) lower and upper ranges

• Important features included: Blur with 4 techniques including Gaussian Blur and Median blur, canny edge detection, erosion & dilation which also includes opening and closing totaling to 13 features

• Reduces hours of work & effort to get precise values for best environment without any hassle

• Works on live stream through android camera, video, images and webcam

• Install by “pip install hsv-calibration” (PyPi.org https://pypi.org/project/hsv-calibration/) Extra-Curricular Activities, Awards and Achievements Smart India Hackathon - National Level Hackathon

• Developed a website in 36 hours for Self-Help-Groups (SHG) in India to expose to the tech-world

• Implemented an e-commerce website for SHGs to sell products at high profits compared to traditional selling by eliminating middlemen. This leads to purging of commission & directly sell it to the customer

• Won 4th Prize all over India, “Best Project” at IIT Bombay and “Best Speaker” at AICTE Conference Prerna Social Welfare Foundation (NGO, Mumbai)

• Raised money for the event by advertisements and sponsors

• Voluntarily developed website for the event registration, 5000+ registration through our website

Contact this candidate