Case Studies | GovTech

Moving an AI exam-grading workload onto AWS

AI Workload Migration

About

A GovTech business working with Cloud Combinator on AWS. The client is anonymised at their request.

Challenge

The AI migration had three focus areas, each of which shaped the target architecture.

Migrating both a live backend and a batch pipeline

The platform backend needed real-time behaviour including websockets, which ruled out some simpler compute options, so it had to move onto container compute that supports long-lived connections. Separately, the exam-grading workload is a bulk-batch job that needed its own managed batch compute rather than sharing the live path.

Consolidating multiple model providers onto one AWS-native service

AI calls were spread across external providers such as Gemini, OpenAI and Claude, including analysis that ran

Absorbing exam-season spikes

Token volumes rise roughly ten-fold during exam periods, so the foundation had to scale up for those peaks and back down again afterwards, and the cost had to be understood at peak sizing so there were no surprises.

Solution

Cloud Combinator structured the work in phases, starting with an assessment that produced a costed plan, then moving the compute onto managed AWS services and the model calls onto Amazon Bedrock before hardening and handing over.

3 to 1

External providers (Gemini, OpenAI, Claude) consolidated onto Amazon Bedrock

10x

Token-volume headroom sized for exam-season peaks (projected)

$195k

Indicative annual AWS run rate at peak sizing, both environments (projected)

By the numbers:

  • 3 to 1 - External providers (Gemini, OpenAI, Claude) consolidated onto Amazon Bedrock
  • 10x - Token-volume headroom sized for exam-season peaks (projected)
  • $195k - Indicative annual AWS run rate at peak sizing, both environments (projected)
Changes

Acceptance was met when the essential AI services ran on AWS as they had on Heroku, with the language-model invocations migrated to Amazon Bedrock and their output validated, best-practice foundations in place, and the environment demonstrated as suitable for the client's continued development. The assessment phase delivered the costed calculator and plan, and the migrate phase delivered the working AI workload and a technical handover.

  • Live backend on managed containers — The platform backend runs on Amazon ECS with Fargate behind an ALB, supporting the websockets the application needs.
  • Batch grading on its own compute — The bulk exam-grading pipeline runs on AWS Batch, replacing a manual upload-and-download process and keeping heavy load off the live path.
  • One AWS-native model service — Language-model calls are consolidated onto Amazon Bedrock, with model access enabled, quota managed, and prompts rewritten for validated output.
  • Sized for the spikes — The foundation and the cost estimate are sized for the roughly ten-fold rise in token volume during exam periods, so peaks are planned for rather than a surprise.
  • Handover — Infrastructure defined in CloudFormation, plus documentation and a knowledge-transfer session, leave the client's team able to maintain and iterate on the AI pipeline.

With the AI workload running on AWS and consolidated onto Amazon Bedrock, the client is positioned to tune prompts and models on a single service, to make further improvements on top of a managed foundation, and to scale both environments through exam-season peaks as demand grows.

AWS Stack

Amazon ECS

With Fargate for running the containerised the platform backend, including websocket support.

Amazon Bedrock

For AWS-native language-model invocation, replacing the external providers.

AWS Batch

For the bulk exam-grading pipeline as managed, scale-to-demand batch compute.

Amazon ECR

For storing the container images, and Application Load Balancer for routing to the backend.

AWS CloudFormation

For infrastructure as code, so both environments are repeatable.

YOU MIGHT LIKE

Related success stories

View all case studies

Case Studies | Insights

Utilising Language Recognition, Speed, and Enhanced Security to Make Social Media a Force for Good

  • Here, we take a detailed look at how the Cloud Combinator team collaborated with another cutting-edge AI service provider that provides intelligent systems to “make social media more social” for brands and users alike.
  • Arwen AI is a UK-based startup specialising in AI solutions to manage and enhance brands’ social media interactions. Founded in 2020 by Matt McGrory, Dr. David Cole, and Joel Bailey, Arwen. AI focuses on using AI to automatically detect and remove spam, toxic comments, and other unwanted content from social media platforms.
  • The team at Arwen have three core products. ‘Moderate’ is focused on identifying and removing toxic content from social media channels. ‘Engage’ helps brands identify and engage with meaningful conversations on social media, and ‘Customize’ allows brands to apply bespoke algorithms to their channels - creating an even more effective moderation and engagement.
Read more
CONTACT US

Ready to turn AI into impact?

We'll help you spot the highest-value opportunities, reduce risk around your first AI initiative, and define a clear path to results from day one.

Why talk to us:

Outcome-driven recommendations

AWS-recognised delivery expertise

Risk-aware AI adoption

Clear next step, not a sales pitch

Start with a focused 20-minute conversation about your goals — no pressure, no commitment.

We use cookies

This website uses cookies to enhance user experience and to analyze performance and traffic on our website.

See our Privacy Policy for details.