Nathan Smith

Nathan Smith

Data Scientist

TerraSense Analytics Ltd

Biography

Data Scientist / Machine Learning Engineer currently working with the TerraSense Team on edge computer vision applications to support mission intelligence. Experience in leading teams to find solutions and working with stakeholders to understand project requirements and communicate results. Skill set in data management and wrangling, statistical analysis, machine learning / deep learning, and data communication. Always interested to learn and apply new approaches and technologies to solve problems.

Previous experience includes a career as a senior water resource engineer leading multi-discipline teams. The role included guiding and managing large sensor based data collection programs to develop analyses supporting design and business decisions. In-house analysis and visualization applications were also developed to automate the growing volume of data collected on projects and provide interactive tools to better communicate results.

A collection of small side project posts and apps can also be found below.

Interests

  • Data Science
  • Deep Learning
  • Computer Vision
  • Edge Computing

Education

  • Master of Data Science, 2021

    University of British Columbia (Okanagan)

  • Bachelor of Applied Science, 2008

    University of British Columbia (Vancouver)

Skills

Python

R

Statistics and Machine Learning

Interactive Data Tools

Project Management

Deep Learning Computer Vision

Posts

Sensor Anomaly Detection

Approach for open-source near real-time anomaly detection with InfluxDB time series database using a LSTM neural network

Log Anomaly Detection

HDFS log anomaly detector with a CNN on feature matrices generated using TF-IDF on log events

Board Game Data Explorer

Board game data explorer app built using Plotly Dash and deployed on Heroku

caRecall R Package

R Package API Wrapper for the Government of Canada Vehicle Recalls Database published on CRAN

ShinyCFA App

Shiny R application hosted on AWS to explore Water Survey of Canada hydrometric data and conduct data summarization and flood frequency analysis

ShinyWeatherCan App

Shiny R application hosted on both AWS and shinyapps.io to bulk download Environment Canada climate data and view summaries of available/missing data

StreamflowPyML

Exploration of streamflow estimation in British Columbia using machine learning

Experience

 
 
 
 
 

Data Scientist

TerraSense Analytics Ltd

Oct 2021 – Present BC, Canada

Development of novel deep learning computer vision applications to support missions intelligence using PyTorch and NVIDIA TAO. Computer vision applications include image classification, object detection, and instance segmentation.

Deployment of applications to edge devices using NVIDIA Deepstream.

Data cleaning and wrangling supporting dataset generation for various sensor modalities.

 
 
 
 
 

Data Scientist - UBC MDS Data Science Capstone

Urban Data Lab

May 2021 – Jun 2021 Vancouver, Canada
Developed a near real-time anomaly detection approach for building sensors using open-source software with InfluxDB including a dashboard and notification system. Anomaly detection used a long short-term memory (LSTM) recurrent neural network with an encoder-decoder architecture model framework in Keras with TensorFlow. The framework was implemented in a test InfluxDB environment with Docker.
 
 
 
 
 

Data Scientist and Senior Engineer

Knight Piesold Ltd

Sep 2009 – Sep 2021 Vancouver, Canada

Data Collection and Analysis:

  • Data collection and management of large data collection programs with projects consisting of up to 50 stream gauges and multi-year timelines.
  • Analysis of energy and hydrometric data to support financial modelling, project design, and permitting requirements.
  • Time series modelling, extreme value analysis, and long-term synthetic data generation using statistical and machine learning techniques.

Development of In-House Applications:

  • Interactive data visualization and analysis applications built in R and Python.
  • Applications developed to reduce time spent on data processing, improve quality control of analyses, and better communicate results and uncertainties to project teams and clients.
  • Technical support and user interface feedback for software development teams working on in-house data management software.

Probabilistic Balance Modelling:

  • Development of water balances and energy models to assess potential future conditions supporting decision making on mining and hydropower projects.
  • Application of deterministic scenario-based assessments and probabilistic sampling methods.

Project Management:

  • Managed multi-year projects with annual budgets of up to $1M.
  • Budget and schedule control, client communication, multi-discipline team management of up to 10 staff, technical reviews, and mentoring of junior staff.

Apps

Board Game Data Explorer
ShinyCFA
ShinyWeatherCan

Contact

  • BC, Canada