Conventional Learning Management Systems (LMS) fall short in delivering adaptive, learner-centric experiences. MAS identified a transformative opportunity: to engineer a human-centered learning platform that could serve as an intelligent AI tutor, dynamically tailoring content and support based on each learner's unique journey. This ambitious vision integrates lifelong learning with holistic well-being, all while ensuring scalability and cost-effectiveness.
The project team, collaborating across continents and organisations, includes experts in Learning Design (USA), Data (USA, C3L University of South Australia, Comunet), front end development (Robots and Pencils US/Canada), and backend development and DevOps (Comunet). Within this organizational framework, Comunet was required to tackle a series of complex and sophisticated challenges in highly emergent technical domains:
- Engineering a progressive agentic AI architecture that maintains context alignment with curriculum requirements, past conversations, and personalisation across multiple agents.
- Implementing a Knowledge Graph for Curriculum extensible for future expansion
- Implementing an efficient memory system specific to the requirements of personalization and AI conversation continuity
- Building a secure and compliant data pipeline for multiple regions
- Optimizing costs across multi-cloud deployments
Partnering with Comunet, MAS implemented a groundbreaking solution anchored by Amazon Bedrock’s advanced AI capabilities. The platform harnesses multiple foundation models through Bedrock's API and other AI providers, enabling:
1. Progressive Agentic AI Architecture:
- Implemented Claude (Anthropic) through Amazon Bedrock for maintaining learning session contexts to pass between Agents. This allowed a smooth handover between discrete Agents while maintaining the perception of the AI being continuous conversation.
- Utilized RAG techniques to enhance the model's ability to reference curriculum materials accurately.
2. Multi-Agent Architecture:
- Developed a hierarchical prompt structure using Bedrock's Claude and Titan models.
- Integrated with LangFuse to enable a separate Prompt Engineering pipeline loosely coupled with the back-end APIs for faster iteration cycles and maximizing knowledge of Learning Design experts and Developers.
- Created an inheritance-based prompt system ensuring consistent AI ‘voice’ and safety protocols.
- Implemented few-shot, ReAct and other prompting techniques to optimize agent responses for educational contexts.
- Created a dynamic multi-agent orchestration flow to enable personalized learning journeys to adapt to preferences and needs of individual learners.
3. Knowledge Graph:
- Co-Designed and implemented a Knowledge Graph for Curriculum in Amazon Neptune.
- System allowed for both a standard path through curriculum and personalized learner journeys.
4. Memory System:
- Built a custom memory implementation using Amazon OpenSearch and DynamoDB for persistent storage.
- Integrated Bedrock's context window management to maintain coherent learning sessions.
5. Event-driven Serverless Architecture and Data Pipelines:
- Established Enterprise-grade Cloud Infrastructure delivered through DevOps as Infrastructure as Code (IaC).
- Real-time data enrichment to enable an ongoing data feedback loop that instantly converts raw learner activity into meaningful, actionable insights.
- Implemented an AWS Lake Formation Catalog to enable synergies between a globally distributed development and data team.
- Implemented data anonymization and regional compliance measures to comply with US and EU legislation and requirements.
- Created efficient data aggregation systems for analytics and personalization.
- Security and Compliance Framework providing deep protections on Learner's data throughout the application and data layers.
The solution integrates with front-end development by Robots and Pencils (USA/Canada), creating a seamless user experience that adapts to individual learning patterns and needs.
Currently in private testing, the MAS platform is poised for commercial launch in early 2026. Early feedback highlights several technical breakthroughs have already been realised:
- The successful deployment of a pioneering AI-powered educational architecture, strategic cost optimization, and robust data management meeting global compliance standards.
- The platform’s scalable, personalized learning experience has garnered enthusiastic responses from initial users, underscoring its potential to reshape the EdTech landscape.
- Successful implementation of a first-of-its-kind software pattern for AI-powered educational platforms.
- Efficient cost management through strategic model selection and multi-cloud optimization.
- Robust data handling system meeting international compliance standards.
- Creation of a scalable, personalized learning experience.
- Successful public testing phase demonstrating platform viability.
- Positive feedback from early users regarding the personalized learning experience.
While quantitative results from the private testing phase are forthcoming, MAS is strategically positioned to drive transformative change in educational technology. By leveraging Amazon Bedrock for AI-driven curriculum, MAS introduces a new paradigm, merging personalised learning with holistic wellness. This case study exemplifies how generative AI, combined with global collaboration and pedagogical expertise, can revolutionize educational experiences and outcomes.
[Note: This case study is based on a pre-launch product, with specific metrics to be determined post-commercial launch in 2026.]
