AI modules in a client's business management and accounting systems
We built AI modules that run in a client's business management and accounting systems: invoice recognition, a proposal generator, assistants for accountants and reports, and MCP servers.
The assistant carries out operations only after the user approves them, and every approval is logged.
- Client
- Software company
- Our role
- New modules in an existing system
- Years
- 2025–2026
- Technologies
- Language models (tool use, structured output), document recognition, MCP, OAuth 2.0
The task
Build language models into a client's business management and accounting systems as a working part of the product, on real business data.
The solution
Modules we built
- Invoice recognition in both systems: a PDF, photo or email attachment becomes a structured invoice, with its lines matched to the catalogue and suppliers.
- Incoming mail processing: attachments from the mailbox go to recognition automatically.
- A proposal generator: proposals are drafted from the customer's data, pricing rules and the product catalogue, and reviewed before they go out.
- An AI assistant for accountants: it helps classify documents, prepare entries and check data, carries out operations only after the user approves them, and its AI use is metered and monitored.
- A reports assistant: it turns a question in plain words into a database query and returns the result, with read-only access.
- MCP servers for both systems: through them, AI assistants work with the data within the user's permissions.
Accounting system
Results
- The modules run in both systems and are still being developed.
- Each time a user approves one of the assistant's operations, the approval is logged.
- The MCP server has 42 tools; sign-in is through OAuth, with the same permissions as the user interface.
Related services and solutions
Want a similar solution?
A few sentences are enough – we'll reply and talk through how we could help.