The challenge
The client runs a project-management platform for agencies that organize corporate events, conferences, and business travel. Managers on the platform build a cost estimate for every event, and the work is manual: dig through past projects, copy the services, adapt them to new dates, participant counts, and durations. Assembling one budget took up to 3 hours.
The estimate is also where deals are won or lost. A manager who answers a client request in an hour beats one who answers tomorrow.
The approach
The assistant lives inside the platform as a sidebar panel, next to the budget the manager is working on. It already knows the active project: the brief, the dates, the participant count. The manager describes what the event needs in plain words, and the assistant pulls matching services from the catalogue and from past projects, adapts prices and quantities to the new parameters, and fills in the budget lines.
The manager stays in control of the result. They review the draft, ask for changes in the same chat, and the assistant edits the budget lines directly.
Architecture
- Backend: Python, FastAPI
- AI: an LLM agent with tools for catalogue search, past-project lookup, and budget editing
- Integration: a REST API the host platform embeds, with search, fetch, and CRUD operations on budget line items
The assistant is built to disappear into the host product. It exposes its capabilities through the API, the platform renders the chat in its own UI, and managers never leave the tool they already work in.
Results
A budget that took up to 3 hours of copying between projects now comes together in minutes of conversation. The manager’s time goes into checking the numbers and talking to the client, and the assistant handles the lookup and data entry.