Machine Learning · Bayes · Robotics

Patrick Dallaire, PhDApplied Scientist

I build intelligent systems that operate in the real world

Currently building

ivr.studio

Design a company's phone greeting — menus, schedules, messages — produce the voices, synthetic or recorded by a real narrator, and deliver a ready-to-install audio package to the technician. No back-and-forth.

Product · in development

Ringuo

A phone number answered 24/7 by an AI voice receptionist: it takes messages, answers customer questions and transfers the calls that matter. For businesses whose hands are busy. Bilingual EN/FR.

About

I am a specialist in machine learning, Bayesian statistics, and robotics. I am the founder of NQB, an artificial intelligence consulting firm that helps businesses solve complex problems through custom AI solutions. I work directly with clients to oversee the definition and implementation of their AI-driven solutions, ensuring that technical development remains aligned with their business objectives. I also lead NQB's product development, where we are building Ringuo, an AI voice receptionist that answers calls 24/7 on our own PBX infrastructure, and ivr.studio, a phone-greeting design studio — engineered for low-latency audio, real-time conversational intelligence, and seamless integration with enterprise telephony. Years of fundamental research have given me a deep mathematical foundation that I bring to every problem I tackle — the ability to see beyond standard approaches, formulate original models, and design algorithms tailored to challenges that off-the-shelf methods simply cannot solve. My research has been published at major conferences including UAI, AAAI, and NeurIPS, and has been widely cited by the scientific community. I have delivered technologies that are now in operation across industries, from industrial inspection and manufacturing to life sciences and food processing, for organizations such as Medicago, Eddyfi, Thales, Optel, and Frontmatec.

« Whether the input is a spectrum, an image, a waveform, or a point cloud, I approach every problem the same way: understand the data, formulate the mathematics, and build an algorithm that solves it. »

Fields of Application09

Hyperspectral Imaging
Mass Spectrometry
Eddy Current Signals
Inertial Measurements
3D Laser Profilometry
Computer Vision
Scintillation Dosimetry
Electroretinography
Structured Data

Technical Expertise09

Deep Learning
Computer Vision
Signal Processing
Bayesian Inference
State Estimation
Unsupervised Learning
Reinforcement Learning
Time Series Forecasting
Optimization

Experience

2021 — Present

Chief AI Scientist

NQB.AI

Leading an AI development firm focused on applying machine learning technologies to solve industrial problems. Business development, technical direction, and delivering cutting-edge AI solutions to clients across various industries.

PythonPyTorchDeep LearningComputer Vision

2020 — 2026

Adjunct Professor

Université Laval

Conducting research in Bayesian machine learning and neural networks. Supervising graduate students in machine learning research projects.

Bayesian MLResearchNeural NetworksTeaching

2019 — 2022

Chief AI Scientist

SmartyfAI

Business development, marketing and administration of an AI development firm. Applying machine learning technologies to solve industrial problems.

PythonMachine LearningBusiness Development

2017 — 2020

Data Scientist

Centre de Recherche en Données Massives

Led research projects and supervised students in machine learning applications. Developed corporate training solutions and advanced data analytics pipelines.

PythonMATLABKerasscikit-learn

2016 — 2017

Computer Vision Researcher

Institut National d'Optique

Developed calibration systems for hyperspectral cameras and laser profilometers. Built classification models for hyperspectral imaging and multi-object tracking algorithms.

Computer VisionHyperspectral ImagingMATLABBayesian Filters

2014

Software Engineer Intern

Yelp

Maintained and developed Yelp ads software, analyzing big data for click-through rate prediction. Built a learning/testing pipeline service for ad targeting optimization.

PythonBig DataAWSA/B Testing

Selected publications

2020

The Indian Chefs Process

UAI · Conference on Uncertainty in Artificial Intelligence

P. Dallaire, L. Ambrogioni, L. Trottier, U. Güçlü, M. Hinne, P. Giguère

2018

Bayesian Nonparametric Deep Learning

NeurIPS · Bayesian Deep Learning Workshop

P. Dallaire, L. Trottier, P. Giguère, B. Chaib-draa, F. Laviolette

2014

Autonomous tactile perception: a combined improved sensing and Bayesian nonparametric approach

Robotics and Autonomous Systems

P. Dallaire, P. Giguère, D. Émond, B. Chaib-draa

2011

An approximate inference with Gaussian process to latent functions from uncertain data

Neurocomputing

P. Dallaire, C. Besse, B. Chaib-draa

2009

Bayesian reinforcement learning in continuous POMDPs with Gaussian processes

IROS · IEEE/RSJ International Conference on Intelligent Robots and Systems

P. Dallaire, C. Besse, S. Ross, B. Chaib-draa

All publications on Google Scholar

Projects

Showing 22 / 22

AI-Driven Robotic Meat Cutting

Developed an advanced 3D vision solution using laser profilometry to extract critical landmarks on meat. Deep neural networks for semantic segmentation and detection enable precise robotic cutting paths.

PythonC++Deep LearningComputer Vision

Non-Destructive Testing with Eddy Current

Built a machine learning pipeline to predict structures in metal components using non-destructive measurement devices. State-of-the-art deep learning with Bayesian inference, deployed via ONNX.

PythonPyTorchONNXMLflow

LCMS Deep Learning for Metabolomics

Predicted plant yield from temporal metabolomics data measured with LCMS. Developed the Reference Concentration Network (RCN) and VectorizedLinear layer running millions of networks in parallel.

PythonDeep LearningSignal Processing

Digital Fingerprinting for Supply Chain

Created unique digital signatures from aluminum product images for reliable object re-identification. Robust matching pipeline effective under varying lighting and perspective conditions.

PythonComputer VisionFeature Extraction

Avionic Bay Temperature Forecasting

Forecasted temperature fluctuations in airplane avionic bays for Bombardier. Created a new Bayesian Variational Inference algorithm to estimate model uncertainty using RNN and GRU architectures.

PythonPyTorchPyroBayesian ML

COVID-19 Wave Forecasting

Combined viral load data from wastewater with epidemiological metrics to detect COVID-19 waves in advance. Integrated time-series methods and deep learning with multi-head attention.

PythonTime SeriesRNNDeep Learning

DEEL — AI Certification Research Program

Conceived a $5.9M research program with 4 industrial partners (Bombardier, Thales, CAE, Bell Helicopter) across 5 universities focused on AI certification. Defined 4 research axes: Robustness, Interpretability, Confidentiality and Certification. Coordinated sub-teams and led the ML literature review.

Machine LearningResearchAI Certification

Predicting House Acquisitions — Desjardins

Anticipated customer mortgage needs based on transaction history for Desjardins. Supervised a team of 4, selected architectures (TCN, LSTM, RNN), evaluated model effectiveness with LIFT metrics and delivered the solution on Azure.

PyTorchLSTMRNNAzure

Student Population Dropout Prediction — Université Laval

Predicted student dropout probability per semester for Université Laval (VREX). Supervised a team of 2, designed the neural network architecture, evaluated model effectiveness and assisted in technology transfer. Built with Keras/TensorFlow and scikit-learn.

KerasTensorFlowscikit-learnDeep Learning

Airplane Tracking with RGB Camera

Developed a tracking algorithm for airplanes using RGB cameras at INO. Combined Gaussian processes for temporal smoothing to handle occlusion with a convolutional neural network for airplane detection. Responsible for implementation, integration and performance validation.

CaffeMATLABGaussian ProcessesComputer Vision

Mushroom Ripeness Prediction with Hyperspectral Imaging

Predicted mushroom ripeness levels using hyperspectral imaging at INO. Developed data acquisition procedures, camera calibration, image segmentation and annotation systems, and a proprietary classifier for hyperspectral images. A patent was submitted for this technology.

MATLABHyperspectral ImagingComputer Vision

Basil Water Needs Prediction with Hyperspectral Imaging

Predicted water needs of basil plants using hyperspectral imaging at INO. Developed data acquisition, camera calibration, image segmentation and annotation systems, and a proprietary classifier for hyperspectral images.

MATLABHyperspectral ImagingComputer Vision

Metro360 — 3D Pipeline Reconstruction

Reconstructed 3D models of pipelines using a robotized laser profilometer at INO. Developed a proprietary Bayesian filter (Extended Kalman Filter) for inertial measurements, data imputation techniques for 3D laser data, and a learning algorithm for spatial-temporal positioning.

MATLABBayesian Filtering3D ReconstructionLaser Profilometry

Mental Illness Detection from Signal Analysis

Developed unsupervised learning algorithms for detecting mental health conditions from proprietary signal data for diaMentis. Created a novel machine learning algorithm tailored to the project needs, implemented as a package for their proprietary ML platform.

MATLABUnsupervised LearningSignal Processing

Collaborative Adaptive Cruise Control

Developed a complete neural network library in C++ for integration into a reinforcement learning-based adaptive cruise control at Université Laval. Built backpropagation, gradient descent, fully connected layers, activation functions and loss functions from scratch using Boost and Lapack.

C++BoostLapackReinforcement Learning

Artificial Finger — Robotic Texture Detection

Detected textures using a robotic finger with IMU signals at Université Laval. Developed a novel unsupervised learning algorithm (DPMoG) to identify textures without labels, and a Bayesian classifier to evaluate performance across 28 test textures.

MATLABUnsupervised LearningBayesian MLRobotics

DEPLUMP — Probabilistic Text Compression

Developed a Bayesian non-parametric model based on the Pitman-Yor process to predict letter and word sequences in text documents at Université Laval. The probabilistic predictor was used to compress documents, packaged as a substitute to rar/zip. Evaluated on the Calgary corpus.

C++Bayesian Non-ParametricsNLPCompression

Walking Robot Surface Detection

Detected surfaces using an IMU mounted on a walking robot foot at Université Laval. Developed a novel unsupervised learning algorithm (PYPMoG) capable of identifying different surfaces without labels, and a Bayesian classifier to evaluate performance across 12 test surfaces.

MATLABUnsupervised LearningBayesian MLRobotics

Indian Chefs Process — Bayesian Network Structure Learning

Constructed a new probability distribution on infinite directed acyclic graphs at Université Laval. Derived mathematical models and Bayesian inference equations for Gibbs sampling and MCMC to learn the structure of neural networks and Bayesian networks. Supervised a research assistant.

MATLABBayesian Non-ParametricsMCMCGibbs Sampling

Drone Relative Positioning

Estimated relative positioning of two in-flight drones using mounted cameras at Université Laval. Implemented fiducial marker extraction, a Particle filter for position estimation, drone dynamics learning, and future position prediction. Built robotic remote control and communication with ROS.

PythonOpenCVROSBayesian Filtering

Ads Targeting — Yelp

Implemented data management and analytics services on AWS to target users with relevant ads. Developed a Bayesian network estimating click probability of users on ads. Integrated models in production, monitored performance, designed A/B tests and wrote unit tests.

PythonAWSBayesian NetworksA/B Testing

AEC Program in Artificial Intelligence — CÉGEP Ste-Foy

Led the scientific design of a new CÉGEP program in artificial intelligence. Defined the skill set, course subjects and material, technologies to be taught, and anticipated the workforce needs of the machine learning industry.

AICurriculum DesignEducation