LLMs in production · agent orchestration · documents · process automation
Site Visit Minutes
A bot turns voice notes and photographs from a site visit into an editable report draft, reviewed by the professional before use.
- Role
- Complete design and implementation
- Period
- September 2026
- Status
- in progress
- Visibility
- Case study
Private repository. It contains minutes and photographs from a real construction site and client.
The problem
After a site visit, the professional needs to turn observations into a report. Voice notes and photographs contain the information, but organising and writing it adds work.
What I built
A Telegram bot collects the material and produces a Word draft. Transcription and LLM extraction propose structured information; the system validates it before writing the document.
Human review
The professional reviews, edits and signs the report. Corrections preserve the history and the workflow limits model calls. Draft quality is assessed through how much content survives editing.
This is a pilot with private client material. The image is an illustrative example without client data.
Screenshots
