Microsoft

Key meetings often ended with hours of manual follow-up. Someone had to review the transcript, determine action points, and create a slide deck — work that was repetitive yet critical to momentum.

AI can summarise content quickly, but its free-form outputs rarely match the strict formatting needed for executive presentations. The challenge was to blend AI’s generative power with a fixed, predictable slide structure that Microsoft teams could rely on.

Security and compliance requirements necessitated that the entire tool run on Microsoft Azure. This added constraints on service selection, architecture design, and performance optimisation — all while ensuring the system could be maintained internally after delivery.

The tool needed to process transcripts in a variety of formats — including .doc, .docx, and plain text. This allowed users to work directly with files from common word processors or simple text exports, ensuring compatibility regardless of how the transcripts were created or shared.

We had 6–8 weeks to research, design, build, and test the MVP, with no room for scope creep.

We designed an architecture that used only Azure services, ensuring seamless deployment and integration with Microsoft Teams workflows.

At the core, the tool uses the Azure OpenAI API to read and interpret meeting transcripts. To keep the output consistent, we designed precise prompts and layered in programmatic rules, which segmented the AI’s responses into discrete slide-ready sections.

Using Node.js libraries in combination with Nuxt.js, the tool converts the structured output into a basic PowerPoint deck based on a predefined slide template. This “bare-bones” version is intentional — it allows Microsoft teams to apply their internal branding and styling without reformatting the content.

A key feature of the MVP was the ability to regenerate specific slides with additional hints. If a section was generated incorrectly, users could refine the output without needing to reprocess the entire transcript.

Testing was run jointly by our development team, our PM team, and Johnnie Studio’s strategy group. We used real meeting transcripts for accuracy checks, gathered immediate feedback, and refined the AI’s prompt and output rules to reduce ambiguity and improve slide quality.

The delivered MVP achieved the core goal while laying the groundwork for future capabilities, such as expanded file type support and direct integration with branded templates.

The project demonstrated that Microsoft’s meeting-to-summary workflow could be condensed from hours of manual work to just minutes of automated processing — without sacrificing accuracy or structure.

Teams can now:

• Upload a transcript immediately after a meeting.
• Receive a consistent, structured to-do list and summary deck almost instantly.
• Share or adapt the deck for executive review with minimal extra work.

Key achievements included:

• Delivered a working MVP in just 8 weeks.
• Fully hosted and integrated within the Microsoft Azure cloud environment.
• Hybrid AI + rules-based method ensured outputs matched strict corporate structure.
• Reduced manual meeting follow-up from hours to minutes.
• Introduced “regenerate with hints” functionality for more accurate results.