Projets novateurs réalisés

Leveraging Advanced NLP Models and LLMs for Document Set Understanding with Multi-documents semantic relations

Docugami is a document engineering company registered in British Columbia that transforms how businesses create and execute critical business documents. Leveraging breakthrough Natural Language Processing (NLP) and Large Language Models (LLMs) Docugami enhances productivity, compliance, and insight across industries such as finance, law, and business operations.
This project addresses a key challenge or objective for Docugami—improving document set understanding by developing advanced methodologies of understanding, chunking and extracting semantics relations across large collections of similar types of business documents. Traditional approaches struggle with complex document structures, limiting automation and knowledge discovery. Additionally, effective connections and understanding relations between document information— which usually are lost in the documents—plays a vital role in driving business efficiency, streamlining processes, and
unlocking valuable insights within the business sector.
The anticipated benefits for Docugami include enhanced AI-driven document intelligence, leading to more accurate information retrieval, document summarization, and knowledge extraction; and improved ability to create integrated and intelligent business systems, providing better service for small businesses’ or large organizations’ needs. The project also aims to contribute to the open-source community by documenting findings, code sharing, and collaborating on projects related to document understanding using LLMs.

RationalTek : Plateforme d’architecture intégrée en IA

RationalTek : Plateforme d’architecture intégrée en IA
Principales activités du partenaire :
RationalTek vise à rendre les organisations plus performantes en les soutenant dans leur projet de transformation organisationnelle par le biais de la mise en place d’une architecture d’entreprise orientée service (AEOS) innovante et pragmatique.
Problématique :
Le projet global consiste à développer une plateforme d’architecture qui sert à la fois d’outil d’architecture d’entreprise et d’architecture de solution. Cette plateforme repose sur un métamodèle complet qui définit les
concepts et relations clés associées à l’EA et à l’AS. L’objectif principal est d’intégrer un moteur d’intelligence artificielle, basé sur un modèle de langage avancé (LLM), capable d’interagir avec ce métamodèle pour fournir des réponses contextuelles et guider les utilisateurs dans des tâches telles que l’ingénierie avancée (forward engineering) et autres cas d’utilisation IA. Par exemple, un utilisateur pourrait interroger le système sur les “Services Applicatifs” associés à une “Fonction Applicative” spécifique, comme l’audit, et obtenir des propositions précises basées sur le métamodèle.
Avantages escomptés du projet :
La plateforme LANA propose une solution intelligente et intuitive qui facilite la conception et la modélisation des architectures tout en optimisant le suivi des projets. Son adoption permet aux entreprises de réduire le temps consacré à la modélisation et à l’architecture, de gagner grandement en efficacité et en qualité tout en réduisant les coûts de dépassement et le taux d’échec des projets.

Mise au point d’un dispositif de traitement électrochimique de l’eau potable des puits isolés

Ce projet de recherche de 4 mois vise à comprendre les raisons des dysfonctionnements observés sur des systèmes pilotes utilisant une technologie électrochimique pour traiter le sulfure d’hydrogène (H2S) dans l’eau de puits résidentiels isolés. En étudiant deux contextes géologiques différents, nous analyserons les paramètres géochimiques et opérationnels qui influencent l’efficacité du traitement. Le projet inclut des analyses en laboratoire et sur le terrain, des tests électrochimiques, une modélisation prédictive, et le développement d’un guide technique pour les opérateurs. Les résultats permettront d’identifier les seuils critiques de certains composants chimiques et de proposer des solutions concrètes pour améliorer les performances des installations défectueuses. Ces travaux offriront à l’organisation partenaire une meilleure compréhension des facteurs limitants et un protocole optimisé pour la maintenance des électrodes, garantissant ainsi une amélioration rapide des systèmes existants et posant les bases pour des avancées futures dans le traitement du H2S.

Improving the efficiency of the supersonic binary fluid ejector using computational fluid dynamic modeling

The only commercially available technology that directly uses thermal energy to produce cooling is absorption chillers,
which are not economical for small-medium scale buildings (<100,000 sq.ft.) and suffer from serious performance limitations. May-Ruben Thermal Solutions (MRTS) is developing a novel Binary Fluid Ejector (BFE) that will provide a high-performance, economic, scalable, thermally-driven heat pump and refrigeration cycle. Early applications include space cooling/heating for residential and commercial buildings, providing economic savings and GHG reductions. MRTS is currently constructing a laboratory BFE prototype. The proposed project includes research needed to support the development of an alpha prototype which consists of a closed-loop recirculating BFE heat pump system, including 3-D computational fluids dynamic (CFD) to be performed by the candidate. This modeling work will enable the design of geometrically optimized ejectors that maintain their performance when used in applications with higher temperature differences, such as space cooling, where traditional ejectors rapidly loose efficiency.

Integrating comparative genomics and Tn-Seq data to uncover genetic interaction networks in pathogenic bacteria

Antibiotic resistance is a rising threat worldwide, making infections increasingly hard to treat. This project aims to tackle this urgent issue by mapping out the hidden genetic relationships within harmful bacteria, including Escherichia coli and related pathogens. By analyzing vast amounts of genetic data, we will identify critical points—called genetic hubs—essential for bacterial survival. Targeting these genetic hubs could lead to the development of powerful new antibiotics, providing fresh solutions to protect public health and combat resistant infections effectively.

Explaining Graph Machine Learning Models via Tensor Networks: A Bridge to Quantum Computing

This research investigates whether tensor networks can serve as interpretable surrogates for graph neural networks (GNNs). It explores whether tensor networks can approximate the functional behavior of GNNs while offering a more structured and interpretable internal representation. The project aims to quantify the contribution of nodes, edges, and features to predictions through this surrogate representation, enhancing model transparency. It also examines how the extracted tensor structure could inform the design of efficient quantum circuits, leveraging the deep mathematical connections between tensor networks and quantum computation.

Tensor networks are compact, modular, and inherently structured—traits that make them promising candidates for interpretable machine learning. Their alignment with quantum circuit models allows not only for a clearer understanding of classical GNNs but also for porting learned structures into quantum-native architectures. This bridges a key gap between explainability in graph machine learning and practical quantum algorithm design.

By combining RIKEN AIP’s expertise in quantum computing and interpretability with University of Montreal and Mila’s strengths in graph models and machine learning, this collaboration creates a unique opportunity to advance interdisciplinary research. Mila gains exposure to advanced quantum approaches, while RIKEN AIP benefits from insights into graph-based AI, enabling new directions in tensor-based and quantum-inspired model development.

Automation of Doppler Ultrasound using Artificial Intelligence

(1) Activities of partner:
Moonrise Medical is developing an AI-enabled ultrasound-based device to evaluate flow hemodynamics in the peripheral vascular system. They are targeting peripheral artery disease [1] and diabetic foot ulcers [2] to promote improved wound healing and prevent amputations.
(2) Challenges the partner aims to solve:
The main challenge being addressed is how to achieve accurate, early, and immediate diagnosis/characterization of vascular pathology in the pedal vasculature to predict wound healing characteristics using a form factor appropriate to in-clinic use [3]. To address this overall challenge, Moonrise Medical needs to develop an ultrasound-based solution with a low barrier of entry, streamlined workflow, and small form factor. This includes devising automation and assistance technology tailored for the target clinical application so that a non-specialized clinician may use the device to assess vascular health with minimal training.
(3) Anticipated social or economic benefits
Peripheral artery disease has been estimated to have a global prevalence of 5.6% globally, higher in highincome countries [1]. Meanwhile, the prevalence of diabetic foot ulcers was estimated to be 6.3%, higher in North America [2]. Both vascular pathologies ultimately lead to adverse outcomes including limb loss and increased mortality rate [4]. Current clinical standards for assessing vascular perfusion in the lower extremities have been found to be erroneous for patients with diabetes and/or noncompressible vessels [5]. A more accurate, accessible, and early detection solution for pedal vascular health assessment would significantly improve the outcomes for patients with such vascular pathology.

Factorization of Multivariate Polynomials over Algebraic Number Fields with Multiple Extensions

Polynomial factorization is a core problem in Computer Algebra, with significant applications across fields such as coding theory, cryptography, number theory, solving systems of polynomials, and algebraic geometry. This project aims to develop an efficient algorithm for factoring multivariate polynomials over algebraic number fields with multiple extensions, addressing a key computational challenge in modern algebraic systems. The partner organization, Maplesoft, seeks assistance in improving polynomial factorization over algebraic number fields and function fields. Currently, Maple often struggles with long computation times or fails to complete the factorization of polynomials arising in practice. This project will contribute to overcoming these limitations by developing a more efficient factorization algorithm.

Enhancing Learning Experiences through Applied AI and Real-Time Integrations

Artha Learning Inc. is a Canadian company that designs custom eLearning and blended learning solutions for corporate, government, and non-profit clients. We’re currently building AIReady—a powerful yet accessible platform that helps learning teams integrate artificial intelligence into their training content and workflows, without needing deep technical knowledge.
Through this internship, we’re hoping to move several parts of AIReady forward. We want to improve how the platform handles Artificial Intelligence (AI) tasks like summarizing and retrieving information from multiple documents, add voice and chatbot features that work with learning management systems (LMSs), and make it easier for clients to use the system through intuitive dashboards and tools. We also need support in building real-world examples, custom demos, and short how-to videos that help users get the most out of AIReady.
For Artha, this project will directly contribute to making the platform more useful, scalable, and user-friendly. This is expected to add significant revenue to the project, and add new subscribers to the service. In the next three months, with technical updates, we expect up to 10 organizational subscriptions at $5000/year. The project will also support our broader mission—helping Learning and Development professionals adopt emerging technologies to create better, smarter learning experiences. We plan to convert some of the technical documentation into marketing materials, thereby increasing the awareness of our product as well. Overall, this internship will bring fresh technical talent into our team and accelerate the impact of AIReady for organizations across sectors.

Using few demonstration videos to improve RL agent’s one-shot performance

(1) Ocado Technology, a division of Ocado Group, specializes in AI-driven robotics and automated fulfillment solutions for online grocery retailers. The company develops machine learning models, robotic control
systems, and computer vision technologies to improve warehouse automation. As a partner in this project, Ocado will provide mentorship, computational resources, proprietary datasets, and robotic simulation environments, supporting research into reinforcement learning (RL) for robotic manipulation.
(2) A key challenge for Ocado is reducing reliance on costly and complex data collection for training RL agents. Traditional Imitation Learning (IL) methods require large-scale expert demonstrations, limiting scalability.
Additionally, RL-based robotic systems struggle with generalization, requiring extensive retraining. This project will explore whether a few low-overhead demonstration videos can improve RL efficiency, leveraging vision-language models (VLMs) and imitation learning to enhance one-shot learning.
(3) By improving RL efficiency, Ocado can accelerate AI-driven robotic deployment, reducing training costs, manual labor dependency, and operational expenses. This will enhance warehouse automation and scalability. Beyond Ocado, the research contributes to smarter AI-driven automation, benefiting industries such as manufacturing, logistics, and healthcare.

Enhancing Plasma Dynamics Modeling in RF Ion Sources Using Improved PIC Simulations and Refined Cross-Section Data

This study utilizes computational plasma modelling using a Particle-in-Cell computer program to simulate and compare hydrogen and deuterium in a volume-cusp ion source to ascertain why H¯ and D¯ output beams beam production ratio is ~3:1 [3-5]. To the best of our knowledge a comprehensive modelling study does not exist explaining this effect. This project will provide an important example regarding PIC code utility in shedding light on measurable plasma phenomena. Partner D-Pace develops, manufactures and sells ion sources. D-Pace has worked to experimentally improve the output of its D¯ beams (relative to H¯) with limited success. The challenge is to utilize the PIC/MCC computational plasma modelling code to simulate the H¯ and D¯ cases to: (i) better understand the underlying mechanisms in the plasma that contribute to the production difference, and (ii) to ascertain next experimental steps based on simulation that would yield increased relative D¯ production. The benefit to the partner company D-Pace would be improved H¯ and D¯ ion source products (particularly D¯) for medical cyclotrons used to produce radioisotopes for the diagnosis and therapy of cancer.

Large language models and other machine learning methods to advance AMD’s hardware and software capabilities

AMD is a leading innovator in high-performance computing, graphics, and visualization technologies, focusing on gaming, immersive platforms, and data center solutions. The company develops cutting-edge hardware
and software solutions to enhance computing performance across various applications, including artificial intelligence (AI), machine learning (ML), and edge computing. AMD is seeking to apply advances in AI and ML techniques, especially Large Language Models (LLMs) to address various technical development opportunities from internal process innovation to improving its software and hardware capabilities in gaming and video processing capabilities. The research areas AMD are seeking to address in the proposed project falls into the following research themes. Research Themes
1. Code optimization for asynchronous and synchronous parallel execution across its diverse array of processor types (such as CPUs and GPUs).
2. With the rapid advancement of LLMs, deploying state-of-the-art AI models presents significant memory and compute capacity challenges, especially on embedded and client platforms. While cloud-based AI inference is widely used, latency, privacy, and cost constraints make edge-based AI processing increasingly critical. AMD aims to optimize heterogeneous inference for LLMs by efficiently distributing workloads across CPUs, GPUs, and NPUs within its APUs. Increasing difficulty in optimizing code for its diverse array of processor types (such CPUs and GPUs), which presents significant challenges in code optimization for asynchronous and synchronous parallel execution. Deploying LLMs with long context prompts also poses significant challenges in terms of memory and compute. This issue is critical in applications like multimodal LLMs, which involve massive input token sizes.
3. AMD’s strategic goal of enhancing AI capabilities for game development, specifically focusing on intelligent and adaptive non-player character (NPC) behaviors. Traditionally, crafting engaging and realistic non-player character behaviour (NPC) requires extensive manual effort. This project leverages recent advancements in Large Language Models (LLMs) and reinforcement learning (RL) to automate NPC training and scenario creation, significantly reducing development costs and timelines while improving game realism and player engagement through AMD Schola.
4. The User Experience Group at AMD Canada is developing a local chatbot for answering customer questions about AMD’s products. In order to ensure that the chatbot accurately communicates product information, the chatbot must use Retrieval Augmented Generation (RAG) to ground its responses in product data. However, traditional methods that use text chunking and vector embeddings on structured and unstructured data struggle to find dynamic relationships between text chunks and extract context for complex queries. This project aims to explore innovative techniques to extract complex relationships and insights from AMD’s knowledge base to enhance chatbot intelligence. Doing so will improve customer satisfaction by allowing the chatbot to more accurately answer customer questions, potentially increasing sales and decreasing the need for customers to communicate with customer service.
5. In the video processing domain, AMD aims to enhance real-time video upscaling capabilities to improve user experience during video playback and streaming. Although existing solutions like Radeon Super Resolution and FidelityFX Super Resolution (FSR) are present, challenges persist in achieving highquality upscaling and super-resolution for compressed video content without introducing artifacts or latency. In addition, conventional high dynamic range (HDR) imaging, which merges multiple standard dynamic range (SDR) images, is impractical due to high computational costs, increased latency, power consumption, and motion artifacts. Meanwhile, the rising demand for video content necessitates more efficient processing techniques, as existing solutions fail to meet bandwidth requirements.
Successfully developing methods and solutions to the above research themes will advance the hardware and software capabilities in video processing and game development. The successful outcomes will also increase the general efficiency in deploying LLM with respect to memory usage and compute further increasing the capabilities of AMD’s hardware for video processing and game development. The economic benefits for AMD go beyond
gaining operational efficiency and promises broader impacts to the accessibility of high performance computing (HPC) by simplifying the process for creating optimized code for power gaming, immersive platforms and data centres, as well as retaining its leadership position in the semiconductor industry.