Personal details

Waqas P. - Remote data scientist

Waqas P.

Based in: 🇸🇦 Saudi Arabia
Timezone: Riyadh (UTC+3)

Summary

I have experience in a wide range of classical machine learning and more recent deep learning algorithms and have developed machine learning solutions to problems in multiple domains.

My experience primarily comprises developing models for time series predictions and anomaly detection based on signal processing and machine/deep learning algorithms.

Biomedicine, finance and transportation are the major domains I have worked during my academic and professional career.

I have also worked as a freelance developer with local startups in computer vision and financial domains and have developed production-ready deep learning models. Since august 2017, I am working as a data scientist developing data-driven solutions for our clients including include KT, ETRI, Seoul Metropolitan Govt. amongst others.

Work Experience

Senior Data Scientist
Data Streams Inc. | Aug 2017 - Present
SQL
NumPy
Pandas
Scipy
TensorFlow
It is a mid-sized Korean company, specializing in data management & Governance products and solutions. Being their lead data scientist, I have utilized tools from Python's data science stack, relational databases and machine/deep learning to develop data-driven solutions to cater for our clients' needs.

Education

Kyungpook National University
Doctor's degree・Machine learning, Signal processing, Hypothesis testing
Sep 2012 - Feb 2016

Personal Projects

Traffic volume prediction
2018
Python
Deep Learning
TensorFlow
Data from individual IoT sensors was pre-processed into a 5-min sampled time series of average travel time for each road. The pre-processed and the road infrastructure data was stored into an SQLite database. Multiple linear and machine learning algorithms were compared for lookahead predictions and it was found that TensorFlow-based Long short-term memroy networks (RNN) yielded the best performance that reduced the prediction error to less than 50% of no prediction.
Content-based image retrieval
2016
Python
Computer Vision
TensorFlow
(As a freelance for a local fashion startup) ~ A convolutional neural network was trained on clothing items as a multiclass classification problem. After training, the middle layer of the CNN was used for feature extraction for the images. CNN features were normalized and feature hashing was used for convenient querying. For a given query image, its features are extracted and hashed. The images with the same hashed representation as that of query image are returned as similar images. (2016.10 ~ 2016.12, implemented in Python and TensorFlow, under NDA)

Certifications & Awards

Machine Learning DevOps
Udacity | Feb 2022