Video presentation

Target group

The IVADO “Scientist in Residence” program is aimed at startups seeking the scientific expertise of a PhD student in artificial intelligence or data science so as to acquire new skills as well as validate and guide their R&D projects and move them in the right direction.

PhD students taking part in this program acquire hands-on experience as instructors in an entrepreneurial environment that is complementary and relevant to their curriculum. This in-company experience enables them to take concrete action in knowledge transfer from academia to industry, and thus contribute to the growth of Québec startups at the leading edge of technological advances in artificial intelligence.

By R&D project, we mean a strategic development project and not a project aimed at rapidly bringing a product or service to market.

Supported fields

This program is designed to support research in the areas delineated in our CFREF funding proposal: data science in the broad sense, encompassing methodological research in data science (machine learning, operations research, statistics) and its applications in multiple sectors, including our priority sectors (health, transportation and logistics, energy, business, and finance) and any other sector of application (e.g., sociology, physics, linguistics, engineering).

Objectives, roles and responsibilities


The goal of this program is to promote transfer of academic expertise to the entrepreneurial ecosystem. This is achieved through customized training and coaching developed by the candidate to meet the needs of the startup’s R&D project. The candidate acquires experience complementary to their curriculum, while the startup benefits from the knowledge transfer and establishes itself as a scientific leader in its field.

Roles and responsibilities

The customized training will be co-ordinated by IVADO and delivered by a PhD student / instructor specialized in the relevant field, in close collaboration with the startup involved. Following are the roles and responsibilities of the various parties.


  • Co-ordination and management of the call for applications issued to startups;
  • Receipt, evaluation and selection of applications;
  • Communication with and identification of the PhD student / instructors interested in the program and the selected R&D projects;
  • Administrative and financial management of the program;
  • Initial meeting and follow-up meetings during the process to ensure smooth progress and satisfaction of the parties;
  • Post-training evaluation and any necessary adjustments to the process.

PhD student / instructor

  • Identification of the knowledge transfer goals;
  • Identification of the relevant issues to be tackled;
  • Identification of and training in available methods and technology tools;
  • Validation of the proposed methods and tools;
  • Knowledge transfer and coaching;
  • Delivery of the documents developed during the process: state of the art, documentation on the proposed tools and methods, adoption plan, next steps.

Startup (once selected)

  • Identification and mobilization of the R&D team that will lead the project;
  • Identification of relevant issues to be tackled during the training and transparency on already tested solutions;
  • Availability, transparency and openness to the knowledge transfer to be conducted;
  • Before the start of the mandate, payment of a contribution of $1,950 plus taxes.


  • March 1st to March 24, 2023: call for R&D projects issued to startups;
  • Week of March 27, 2023:  analysis and selection of projects;
  • Week of April 3, 2023: communication with the chosen startups;
  • From April 10, 2023: communication and sharing with PhD students;
  • From May 1st, 2023: start of training and coaching.*

*The start date may vary depending on the candidate’s selection date, availability and administrative delays.

*Doctoral students can apply starting from March 1, 2023. See details in the submitting application section.

Project process

Knowledge transfer plan

The knowledge transfer plan will cover the three phases of the program: preparation and startup, training and validation workshops, and the transfer itself. The number of weeks dedicated to each phase and the main tasks indicated may be subject to change depending on the specifics of each project. A preliminary plan will be requested from the PhD student / instructor when they submit their application and a detailed plan must be submitted at the end of the preparation and startup phase.

Phase 1: Preparation and startup

Given that “Scientist in Residence” is a program for transfer of applied knowledge to an R&D project in artificial intelligence specific to each selected startup, the preparation and startup phase is crucial to achieving smooth progress and the satisfaction of all parties.

We estimate that three weeks (45 h) will be required to complete this first phase. The primary tasks will be structured as follows:

Phase 2: Training and validation workshops

This phase forms the core of the program and, as its name suggests, is the period in which the knowledge transfer will be concretely conducted. Eight weeks (120 h) are set aside for this phase, which will allow time to transfer the theoretical and methodological knowledge as well as the relevant tools for the project, test them in real-world situations, obtain feedback, and make any necessary adjustments.

Phase 3: Transfer and coaching

This last phase consists in ensuring adoption of the best methodologies and tools identified, and completing the knowledge transfer as the R&D project progresses. In addition, the PhD student / instructor will support the startup in defining the next stages of the R&D project. Lastly, this is when the PhD student / instructor finalizes documents benefiting the participating startup at the end of the program (e.g., state of the art, theoretical and practical training, adoption plan). Four weeks (60 h) are reserved for this phase.

Amounts, duration and terms of payment

  • The program will last 15 weeks and must be completed between May 1 and August 31, 2023. The PhD student / instructor investment is +/- 15 hours per week, working jointly with the startup.
  • The training is valued at $13,000 and is eligible for the Scale AI reimbursement of 85%.
  • IVADO will invoice the participating startup the amount of $1,950 + taxes, i.e., 15% of the cost of training. The invoice is payable before the start of the project.
  • The PhD student / instructors will receive a $10,000 scholarship, to be administered by the academic supervisor.
  • A maximum of 15 projects will be funded.

Eligibility criteria

For the startup

  • Have less than 20 employees;
  • Have an R&D team active in the submitted project;
  • Be part or have completed a program of a recognized coaching organization;
  • Pay a contribution of $1,950 + taxes to IVADO before the start of the mandate.

For the candidate

  • Be enrolled in a PhD program at a Québec university;
  • Have the support of the thesis director;
  • Neither the supervisor nor the PhD student must have an employment or business relationship with the startup.

Submitting an application

For the candidate

The application package must be sent via e-mail to and include:

  1. A letter of intent outlining the applicant’s interest in the research project and the fit of their expertise with the mandate (max. 1 page);
  2. Résumé in an open-source format;
  3. A letter of support from the supervisor confirming the applicant’s expertise in connection with the research and the supervisor’s support for the initiative; this template may be used.

For the startup

Complete and submit the form before midnight on March 24, 2023. Submissions that are incomplete or sent by any means other than the form will be automatically rejected.

Evaluation of applications

For evaluation of the R&D project applications submitted by the startups, a committee comprising an academic representative, a representative of the entrepreneurship support organizations ecosystem and the IVADO Entrepreneurship Advisor will be formed. Compliance with the eligibility criteria, the quality of the proposed R&D project and the project’s strategic impact on the company will be the main criteria evaluated.

Candidates interested in a project will be interviewed by the company, which will make the final choice.

Selected R&D Projects


Artificial intelligence for soil health

ChrysaLabs is a Montreal-based agricultural technology company that has developed a soil fertility analysis solution for precision agriculture. The developed solution, based on spectroscopy, infonuagic and artificial intelligence, allows agronomists and producers to measure the quality of their soils in just a few seconds and to take relevant actions to improve the situation. The low costs associated with the technology, the ease of deployment, the speed of acquisition and the accuracy encourage agronomists to take more soil analysis to better understand the health and fertility of the soil and apply only the right fertilizers, at the right time, in the right place and in the right quantity. A better spatial knowledge of nutrients leads to better management of chemical inputs and, therefore, to a positive impact on the economy and the environment.

Real-time soil health analysis enables better agronomic and agricultural decisions for crop yields and the environment. ChrysaLabs has developed a handheld probe for real-time estimation of soil characteristics based on spectroscopy and artificial intelligence. The overall goal of ChrysaLabs’ work is to improve machine learning algorithms for soil feature estimation to facilitate continuous and personalized learning in new regions.

Using data from multiple clients in North America, the project aims to determine a semi-supervised continuous learning strategy that limits the need for laboratory-based probe calibration. This project aims to improve the knowledge of machine learning adapted to the reality of our application framework and our business reality (i.e., accurate and reproducible measurement instrument, high soil variability, need for good market penetration, high data cost) and to develop a continuous and semi-supervised learning strategy.

Profile required:

  • Semi-supervised learning,
  • Data representation/encoding,
  • Experience with real data from multiple sources

For more information on the project, please contact


Neuro-A.I. : Temporal optimization of neurostimulation to improve movement after paralysis

NeuralDrive is a medical device company designing technology that enables unprecedented control of neural stimulation delivery. Our primary mission is helping people move again after a paralyzing spinal cord injury. We introduced personalized neurostimulation of the brain’s motor cortex during motor training. We have demonstrated in animal models that this unique technique immediately improves movement, posture, and fosters long-term recovery from paralysis. Our innovative intervention has the potential to profoundly change the way we deliver rehabilitation today, boosting motor training with precise electronic medicine. While using implantable technology that has already been validated for human use, the radical novelty of targeted cortical stimulation can be applied to a range of other motor diseases, from tetraplegia to certain forms of stroke. Moreover, we possess a functional AI stimulation control module which allows efficient optimization of multi-channel neurostimulation.

Complex movements are composed by a sequence of actions. Neurostimulation can help a user who has motor deficits recover control of movement. Supporting movement in most cases requires not be a single stimulus but rather a continuous goal-oriented intervention, to obtain a sequence of motor actions. This project will push our development of neurostimulation control beyond our single-stimulus optimization and into optimization of goal-oriented stimulation sequences. This will allow us to position NeuralDrive as the current most advanced technology for neurostimulation control.

For more information on the project, write to

LS Tech+

Calibration of analog sensors for cycling performance

LS Tech+ designs, manufactures and sells technology for recreational athletes.

The Wattza™, is a power sensor that measures the effort expended by a cyclist by reading the force exerted on the pedals in Watts and the pedaling speed (cadence). It is light, small, efficient and easily transferred from one bike to another. Its portability is a great asset for those who use it recreationally, or for those who use it in multiple situations and on multiple bikes. It is an ideal tool for the cyclist who wants to improve their cycling.

The objective of the R&D project is to improve the accuracy of the device and its repeatability through machine learning techniques. Developing a clear and simple method to characterize the FSR outputs according to the identified variations is very significant for LS Tech in order to make the product more competitive on the market.

To do so, it will be necessary to study the behavior of piezoresistive sensors (FSR A-201 from Tekscan) by capturing a maximum of data. The sensors do not have a linear output at the applied force. Therefore, a rigorous study is required to properly translate their output signal into a consistent value in Newton. It is necessary to synthesize a mathematical characterization of the FSR responses according to the variations at the product rate (the Wattza) and the resistance that a cyclist must overcome.


  • Write and execute test protocols.
  • Research sensor characteristics.
  • Analyze experimental data.
  • Determine limitations of developed FSR calibration methods.

Profile required:

  • Machine learning (time series experience)
  • Python programming language
  • Knowledge in operations research is an asset

For more information on the project, write to


Graph-based Generative Architecture

Maket technologies is a platform that allows architects to facilitate the creation and prototyping of new architectural concepts using artificial intelligence. In a few seconds, Maket enables the architect to generate a wide range of new floor plan concepts based on the client’s needs in a collaborative environment to facilitate communication and presentation of the project to the client. Their clients can easily put annotations on the generated plans and collaborate with the architect in a conceptual phase before  exporting plans into Autocad.

At the moment, Maket allows architects to generate hundreds of floor plans in a few seconds based on a few client criteria. The current result is an image generated via a generative adversarial network (GAN).

Maket would like to develop a complete generative pipeline that adapts to as many constraints as possible to make a floor plan editable and malleable and thus produce plans that exactly meet the architects’ needs. We are currently working on using graph-based approaches that are based on RNN’s, traditionally reserved for molecule generation.

Our goal is to use this technology to improve the efficiency, sustainability and quality of the architectural design process while mitigating the costs and uncertainty that often accumulate during the process.

Specifically, the goal is to be able to direct our research and development team towards plausible hypotheses and maximize our R&D efforts. Knowledge transfer will allow us to test technical hypotheses rapidly. At the moment, our GAN-based output does not allow us to make a plan scalable and does not comprehensively address the physical and regulatory design constraints that the building must have to maximize its efficiency.

Many of our clients have expressed a desire to not only generate an architectural plan using specific constraints, but also to be able to move rooms and expand the project after our algorithm has generated the plan. The architect will be able to define constraints that will be entered into the system and these will be returned in the form of graphics representing a floor plan that respects the constraints provided. If the architect wishes to iterate on a floor plan that the system has produced, they will be able to modify the constraints (add nodes and change node priorities) and then generate a new graph from the base graph taking into consideration the new updated constraints.

Profile required:

  • Graph theory
  • Operations research

For more information on the project, please contact

Acrylic Robotics

Machine learning for fine arts

Acrylic Robotics is a robotics startup based in Montreal on a mission to make fine art accessible to the general public. Our technology lets artists make limited-edition painted replicas of original pieces – like a printing press for painted art.

Our motion capture system collects data about each brushstroke and creates a digital fingerprint file that is sent to our painting robots to bring to life, replaying each stroke layer by layer with real paint on canvas. A machine learning-powered optical feedback loop ensures that the robot corrects its mistakes and continuously improves its technique to best match an artist’s style. We’re part of the Next AI Montreal cohort this summer and have previously participated in Centech (Propulsion), Next 36 and the Creative Destruction Lab (AI stream).

Acrylic is looking into the development of a machine learning solution that can accurately estimate the pose of a paint brush that corresponds to an image of a paint stroke. The proposed project will involve the use of reinforcement learning to combine both time series data and imagery to develop a model that can reverse-engineer the motion path of the paintbrush required to create a target brushstroke. This allows us to transition our data collection method away from expensive optical motion capture cameras to normal cameras, allowing us to capture data from artists around the world at a much lower upfront cost.

Profil required:

  • Reinforcement learning
  • Robotics
  • Computer vision
  • Movement recognition

For more information on the project, please contact


Automatic detection of transport modes

Greenplay aims to simplify your active or sustainable mobility initiatives by introducing reliable metrics, fun challenges, and tangible rewards. Our mission is to inspire communities to adopt sustainable behaviors through innovative, reliable and easy-to-use technology. Above all, we want to put people and innovation at the heart of sustainable mobility initiatives

In this project, we will explore new ways to improve our mode (bus, train, metro, car, walking, running, cycling and carpooling) detection tools through innovative technologies and new processes. The goal is to work with our team of developers to implement exploratory methods to increase the reliability of the algorithm and processes.

Profile required:

  • Machine learning
  • Statistics and data analysis
  • Interest in gamification
  • Good knowledge of geolocation and sensor data from mobile devices
  • Python and Protocol buffer (GTFS)
  • Tensorflow

For more information on the project, please contact


Estimation of magnitude of stochastic flood using a machine learning-based model

Geosapiens specializes in flood modeling and management. We offer web based solutions to help our clients (municipalities, insurers, real estate) to better manage the risk and reduce its impacts.

Return period flows are employed in a variety of sectors, including environmental management and flood planning. Because most basins are ungauged, these processes are frequently carried out without the use of observed quantile flows. In this situation, regionalization algorithms are frequently used to forecast flow distribution using data from gauged basins. The development of regression models that relate observed flows to independent factors derived from physical and meteorological basin data is a frequent sort of regionalization method. Hydrologic models based on estimated parameters and climatic forcing data can be used to calculate percentile flows as an alternative. However, the ease with which regression models may be used to forecast percentile flows in ungauged basins presents an opportunity. For very large research areas with a considerable variance in flow regression models are known to perform poorly. The recent advancement of machine learning model presents us with an opportunity to create a unified model for Canada and the US.

We want to use a machine learning-based model to predict the distribution of the annual maximum discharge for individual catchments in Continental US. Specifically, a set of catchment geographic, topographic, topological, and climatic factors will be used to predict the distribution parameters at the catchment level. Integration of extreme value theory to machine learning and fine tuning the performance of the model to a desirable accuracy, specifically avoidance of mode collapse will be one of the key factors.

The output of this model will be the input to our hydraulics model which, in turn, will generate flood-risk maps. Our core business is flood risk assessment for buildings and this model will directly benefit us. Once developed, our R&D team will continue to improve the model by applying it to new locations such as Europe and Asia, and configuring it for additional inputs.

Profile required:

  • Extreme value theory
  • Machine learning
  • Statistics and data analysis
  • Spatial statistics
  • Python or other programming languages.

For more information on the project, please contact


Clear vision in Minimally Invasive Surgery

Vope Medical delivers constant clear vision in minimally invasive surgery. Vope’s AI-driven software optimizes the lens cleaning process and minimizes disruptions through automation. This allows the surgeon to stay completely focused on the procedures they are performing, eliminating distractions that arise from the current lens cleaning process.

Endoscopic interventions offer many advantages for patients and in the delivery of health care such as faster recovery times and shorter hospital stays. Maintaining a clear vision through the endoscope plays an important role for safe and efficient minimally invasive surgery. During medical interventions, physical contaminations are occurring by bodily fluids including blood, pieces of tissue, condensation in form of fog, and smoke. Currently, the medical procedure will have to be interrupted frequently to physically clean the lens.

In this project, a vision-based approach to detect and remove the contamination during surgical procedure is proposed to reduce the frequency of lens cleaning required. We propose a framework to automatically remove the contamination from the endoscopic lens and defog the surgical view by application of state-of-the-art image defogging and sharpening algorithms. In particular, the system of our partner will detect the contamination of the endoscopic lens during the operating procedure. Once the percentage of contamination exceeds a threshold, the hardware lens cleaning equipment will receive a signal to be activated and will clean the lens automatically to reduce distractions. The surgeons will thus not need to clean it manually. Meanwhile, image processing techniques will be applied to further defog the surgeon’s vision. Fog is caused by condensation on the surface of the lens because of the difference in temperature and humidity within the cavity where the medical procedure takes place.

The PhD student will advise and support the team in:

  • The creation of a pipeline for processing training data, its execution on existing data; this should allow easy use for future data acquisition ;
  • Creating and improving the defogging algorithm according to best practices in the relevant literature and optimizing its execution in a real-time edge device environment;
  • The creation and improvement of the contamination detection algorithm.

Profile required:

  • Computer vision – Image de-noising;
  • Deep learning;
  • Experimental methodology to set up a ML pipeline.

For more information on the project, please contact

Note regarding intellectual property

As this is a training and knowledge transfer program, all intellectual property belongs to the startup. If the successful applicant, jointly with the company, identifies a new development project and the successful applicant’s involvement is sought, that project would then be the subject of a collaborative research agreement distinct from this program.

A non-disclosure agreement (NDA) may be signed by the successful applicant and the company.


  • Pour les doctorant.e.s
    • Faire preuve d’intégrité et de respect dans tous vos échanges avec le startup;
    • Reconnaître le soutien d’IVADO en mentionnant le nom de l’institut dans toute communication publique au sujet du programme et participer, lorsque possible, à ses différentes activités.
  • Pour la startup
    • Fournir un environnement de travail qui convienne à l’achèvement du projet;
    • Faire preuve d’intégrité et de respect dans tous vos échanges avec doctorant.e.
  • Our commitment to equity, diversity and inclusion
    • To ensure all members of society draw equal benefit from the advancement of knowledge and opportunities in digital intelligence, we promote principles of equity, diversity and inclusion across all our programs, and we commit to providing a recruitment process and research setting that are inclusive, non-discriminatory, open, and transparent.