SchoolGPT & Catalogue Search
The Canada School of Public Service's (CSPS) primary function is to provide government employees with critical training and professional development opportunities. However, the existing D2L Course Merchant catalogue struggles to meet modern business needs due to limited content governance, tagging management, and poor search quality. The platform's keyword-based matching lacks semantic understanding, and does not adapt to evolving user behavior. These challenges diminish the learner experience and hinder the organization's ability to reach the desired audience with its offerings. The renewed catalogue search introduces advanced capabilities to improve discoverability and relevance. Natural language processing (NLP) allows users to search with semantic intent rather than just matching words. Product metadata is enriched with additional contextual information, such as related skills, prerequisites, governmental policies, and the actual content within a product. Search ranking is further refined using live UX analytics and feedback mechanisms to improve accuracy over time. A centralized ontology integrated with the data warehouse and AI platform ensures consistency across the organization and improves classification of offerings. SchoolGPT builds upon these enhancements, transforming how users interact with the catalogue by enabling a natural, conversational search experience. Instead of rigid keyword queries, learners can ask nuanced questions like "Which courses can help me prepare for a management role?" or "What training is recommended for employees in cybersecurity?".
Canada School of Public Service
0+
0.000%
< $0
HITL AI Platform
* Over $350,000 in annual savings by replacing a vendor solution, not included as part of OpEx estimate. The rapid adoption of data analytics and AI/ML has created challenges around data governance, ontology consistency, and data integrity. Organizations often operate in silos, leading to duplicated efforts, inconsistent taxonomies, and difficulty in aligning data models with business objectives. Furthermore, the absence of structured collaboration between stakeholders results in data that is either poorly labeled, lacks domain context, or is misaligned with decision-making needs. To address these challenges, the Human-in-the-Loop (HITL) AI Platform provides a structured and collaborative environment where users of various disciplines and skills can contribute to AI development in a controlled and transparent manner. By integrating aspect-based analysis, data labeling, model refinement, and validation workflows, the platform ensures that AI models are not only well-trained but also aligned with policy objectives and operational realities. The platform enforces data standardization through a centrally managed ontology, ensuring that metadata and labels follow department-wide consistency. Governance policies control data access, ensuring that personnel can only make modifications to their respective scopes. Policy analysts and subject matter experts can review and label data, while data scientists and engineers refine models using this high-quality, domain-informed input. Real-time human validation ensures coherence across datasets before models are deployed. By centralizing AI workflows, teams across the organization can collaborate on datasets, reducing duplication and aligning AI efforts with departmental priorities. Data lineage and audit logs provide transparency on changes and fine-grained control over model evolution. The HITL platform is deployed on Azure, leveraging a stack of cloud-native services: - Azure Data Lake, Cosmos DB, and Synapse Analytics serve structured and unstructured data, providing efficient retrieval and processing for model training. - Azure AI Search and OpenAI enable intelligent querying, search augmentation, and rich semantic insights across datasets. - Azure Virtual Machines, DevOps Pipelines, and App Services are used for continuous integration and deployment of updates to the application and data layers. - Role-based and row-based access controls (RBAC) integrated with web applications prevent unauthorized access and modifications.
Canada School of Public Service
0+
0.000%
< $0
Project MapleOps
The COVID-19 pandemic reinforced the need for modern operations centers to rapidly deploy, adapt, and reorganize in highly volatile environments. To meet this challenge, a zero-trust platform was developed to support decentralized operations while maintaining strict access control and data security. The platform consists of a suite of applications, each tailored to specific operational use cases, with access strictly governed by role and need-to-know principles. A central command group sets objectives, defines policy, delegates authority, and monitors operational data in real-time. Senior, junior, and non-comissioned officers manage their respective formations including further delegation, data provisioning, and data collection. As part of zero-trust edge security, the platform enforces invitation-based authentication and asymmetric encryption. Users can only access the system with a uniquely pre-signed invitation issued by their immediate chain of command. During onboarding, each user generates a password, ensuring that neither platform administrators nor external actors can decrypt data without both delegated and user secrets. This also ensures that data can be securely accessed and stored on the local device, even with intermittent network access or in hostile environments where the device could be compromised. On-device data is protected through a layered encryption model, requiring: 1. Valid platform authentication to firstly retrieve encrypted data according to the user's assigned scope. 2. Proximity to the grantor or immediate chain of command who can revoke delegated authority at any time. 3. The user's correct personal password. Development began on Google Cloud, leveraging confidential VMs to ensure code integrity and security. Anthos was used for multicloud replication into the preferred environment prior to a full migration to Azure for production.
Canadian Armed Forces
0+
0.000%
< $0
Passbox
Teams across organizations frequently need to transfer files securely without a reliable means to do so. For example, Outlook and Teams files are subject to Microsoft's data collection and are often saved to OneDrive by default, exposing them to misconfigured RBAC. This solution is a virtually-free self-hosted substitute for Dropbox, Google Drive, or OneDrive which can be deployed and fully data-wiped in under 10 minutes.
Government of Canada, Open Source
< $0
A* Pathfinding
GameMaker's built-in pathfinding uses a 2D grid with simple box checks for collisions. Out-of-the-box, this does not support more complex scenarios including 2.5D or isometric games. To solve this, old modules from personal indie game development were refactored into a library for quick integration and plug-and-play.
Open Source