An AI Platform That Turns Client Meetings Into Finished Reports

The challenge

Financial advisers, insurance advisers and mortgage brokers document every client meeting. A file note, a fact find, a statement of advice: each one gets written after the call, from memory or from notes scribbled while trying to listen. Taking notes during the meeting pulls attention away from the client. Replaying a recording to check what was said takes as long as the meeting itself. Reports land late, and they read differently depending on who wrote them and how long their day had been.

We wanted the path from a booked meeting to a client-ready document to take minutes, with the adviser's time going into reviewing a draft.

What we built

FinPlan, a complete application covering every step between the calendar invite and the client's inbox. Each part has its own case study:

• A Meeting Bot That Follows the Adviser's Calendar: a bot joins every online meeting on the adviser's calendar and records it
• Speaker-Labelled Transcripts From Client Meetings: each voice is separated and named from the meeting invite
• Readable Meeting Transcripts That Keep the Detail: a clean-up pass adds punctuation and paragraphs under strict rules against summarising
• Advice Documents Drafted From the Adviser's Own Templates: reports are generated section by section from templates each adviser designs
• From Draft Report to the Client's Inbox: a rich text editor, PDF and Word export, and email to the client
• On-Demand AI Processing With Usage-Based Plans: recordings wait until they are needed, and credits are only used on success
• Three Specialist Advice Workspaces From One Codebase: separate brands for wealth, insurance and mortgage advice
• Switching AI Providers Without Changing the Product: speech and language models can move between OpenAI and Microsoft Azure
• Testing AI Report Quality Before Advisers See It: an admin workspace for trying templates against real transcripts

Technology

Kotlin and Spring Boot backend with a GraphQL API of 18 queries and 29 mutations, plus REST endpoints for sign-in, billing, transcription and webhooks. MongoDB for storage. A Next.js and React front end with a TipTap editor. Recall.ai runs the meeting bots, Azure AI Speech handles transcription, and GPT-4o and GPT-4o-mini clean up transcripts and draft reports. Stripe handles billing, with Azure Service Bus and SendGrid for email. Everything ships as Docker containers to Azure App Service through GitHub Actions, with separate staging and production environments. We built FinPlan and we run it in production, testing changes in staging before they reach advisers.