25Eight is an Australian company reimagining business education, combining proprietary diagnostics, accredited learning, one-to-one coaching and mentoring. Central to their model is the idea that content should adapt to each learner. While the core concepts stay consistent, the examples and analogies used to explain them should be personalised to each person’s background and interests. Doing that by hand could not scale, so 25Eight engaged Infostatus to build a generative-AI engine that tailors content automatically.

The challenge

Manually creating personalised examples for every learner across many courses was an unsustainable workload. Without an automated engine, 25Eight could not fully realise the engagement and retention benefits of their tailored approach, and content updates were slow to deploy.

What we did

Approach and methodology. We worked in agile sprints with continuous client collaboration, provisioned the entire environment with Terraform, and applied AWS Well-Architected principles with a privacy-first design throughout.

A multi-stage AI pipeline. Orchestrated by AWS Step Functions, the serverless pipeline analyses each learner profile, decomposes and structures the source course content, identifies where to insert personalised analogies, and generates them with Amazon Bedrock using Anthropic’s Claude models. A quality-assurance stage combines automated model-based evaluation with human-in-the-loop review.

AWS services. Amazon Bedrock with Claude, AWS Lambda, Amazon API Gateway, AWS Step Functions, Amazon S3, Amazon DynamoDB, AWS KMS, Amazon CloudWatch and Amazon VPC, with AWS IAM enforcing least-privilege access and all data encrypted at rest and in transit.

Outcomes

• A working proof of concept for AI-personalised content at scale
• A clear path past the manual-personalisation bottleneck
• An unexpected multi-language capability, opening global reach
• A modular, well-architected, handover-ready AWS solution

Lessons learned

Using different Claude models for different tasks balanced quality and cost, prompt versioning treated prompts as first-class assets, and reusable Terraform modules will accelerate future work.

“We’ve had a chance to review it, and we’re excited about where this project is going and the possibilities it opens up! We love and are excited by what you’ve built so far.”

Matt Day, Principal Software Engineer, 25Eight