Building an AI Report-Generation App That Cut Report Time by 98% on AWS Bedrock
30
days to build
98%
faster report generation
0
generation failures after deployment
Key Insights
Location
Estonia
Project duration
~1 month
Industry
EdTech
Technologies used
AWS Bedrock, Amazon S3, IAM, CloudTrail, macOS, GitHub Releases
Services
The client is a consultant working in the EdTech industry. He reviews client cases and advises on the advantages, challenges and potential paths forward for each one.
This kind of need shows up across consulting practices that process large volumes of intake information within a specific industry and want a streamlined way to reach conclusions and produce reports that are easy to read and share.
The Challenge
His core problem was productivity: to grow his business and take on more clients, he needed to generate reports faster.
Before the solution, producing a single report took him at least half a day. He would manually type out a large system prompt, then describe the task in general language or by supplying documents rather than a structured format. Once generated, the report needed to be verified. With larger documents, AI tools could introduce formatting errors, sometimes requiring the report to be regenerated entirely.
The Solution
The team built a macOS application, distributed via GitHub Releases as downloadable app builds. The app includes:
- A Settings area for managing access keys, editing the system prompt and selecting the model.
- An Input & Output interface with a facts input form and a streaming response pane.
- History, listing past reports with search/filter and export as PDF, HTML or JSON.
- A local cache (stored under ~/Library/Application Support) holding cached reports and generated exports.
On the AWS side, the app connects to:
- IAM Access Keys , scoped specifically to Bedrock and S3.
- Bedrock , handling reasoning and streaming responses.
- CloudTrail , providing an audit log of API calls.
- An S3 outputs bucket with default encryption, storing generated outputs.
The flow: the app sends credentials and an augmented prompt to Bedrock via the scoped IAM keys; Bedrock streams a response back to the app; API calls are logged via CloudTrail; outputs are stored in the S3 bucket; the app then retrieves outputs to populate history and local cache.
This is the same foundation we use for AI agents on AWS Bedrock , built around scoped access and an auditable request path rather than a loose integration.
The build took about one month, largely because the client had already provided a clear description of his needs. Most of the time went into fine-tuning system settings rather than building from scratch.
The solution streamlined the process of entering the information needed to generate a report, cutting generation failures down to zero and reducing report generation time to under five minutes.
We've used the same Bedrock and scoped access pattern on other AI builds, including our AI grant management platform on AWS Bedrock .
Lessons Learned
The same prompt and expected output don't transfer across different foundational large language models. When the team attempted to switch models, they discovered that switching isn't as simple as swapping one out for another. All the settings need to be re-fine-tuned to make the new model work with the existing setup.
If you're planning a similar build, our guide on building an AI agent MVP on AWS Bedrock covers how we approach model selection from the start.
Future Steps
If work continued on this case, the team identified two potential next steps:
- Adding monitoring for failures in the application, since it depends on third-party services that can fail for reasons outside their control (e.g., health checks).
- Adding authentication and permission levels, so the application could be used by teams rather than a single person — for example some users only downloading reports while others generate them.
We build working MVPs on AWS even when requirements are still taking shape, so you can test the idea instead of waiting on a perfect brief.
FAQ
About one month. The client had already provided a clear description of his needs, so most of the time went into fine-tuning system settings rather than building from scratch.
Amazon Bedrock handles the AI reasoning and streaming, an S3 bucket with default encryption stores the outputs, IAM access keys scoped to Bedrock and S3 handle access, and CloudTrail logs every API call for an audit trail.
No. The pattern fits any consultant who processes large volumes of intake information within a specific industry and needs a fast way to turn it into a clear, shareable report.
It's not a simple swap. The team learned firsthand on this project that prompts and settings need to be re-fine-tuned for each new model rather than carried over as is.
Yes. IAM access keys are scoped specifically to Bedrock and S3 rather than granted broad account access, and CloudTrail logs every API call so activity can be audited.
Not yet. It's built for a single user today. Adding authentication and permission levels so teams can use it — with some people only downloading reports while others generate them — is one of the next steps identified for this project.
Viktoriia Kotliarova
Quality Assurance Engineer at Perfsys with 3+ years of experience ensuring software reliability and performance across cloud-based systems. She works closely with engineering teams on real-world projects, bringing hands-on testing expertise to every topic she covers.
Recommended for You
AWS Experts, On-Demand
Need to move fast? Our cloud team is ready to scale, secure, and optimize your systems. Get serverless expertise, 24/7 support, and seamless CI/CD pipelines when you need it most.
Please accept cookies to load the booking widget.