Ai-Smeta
Voice note → construction estimate in 60 seconds via Telegram
Construction foremen don't use apps — they use Telegram voice notes. Ai-Smeta meets them where they are: send a voice note describing the job in any language, and the AI agent transcribes it, extracts materials and quantities, applies the firm's private rate memory, and returns a formatted cost estimate for approval — all within Telegram.
The Challenge
Construction estimation is a manual nightmare. Foremen describe jobs verbally. Office staff translate voice notes into spreadsheets. Rate lookups are done from binders. A single estimate takes hours. Code-switching between languages makes automated transcription unreliable.
The Solution
We built a Telegram bot pipeline: voice note → Whisper API (handles multi-language code-switching) → Claude Sonnet agent extracts materials, quantities, and specifications → queries the firm's private PostgreSQL rate database → generates a formatted estimate with line items, unit costs, and totals → returns it in Telegram for one-tap approval or edit. The rate memory learns from corrections.
The Results
Reduces estimate creation from hours to 60 seconds. Handles mixed-language voice input that defeats standard transcription services. Private rate memory means estimates get more accurate with use. Zero behavior change required — workers keep using Telegram voice notes exactly as before.
Tech Stack
Have a similar project in mind?
We build AI-powered products for businesses ready to digitize their operations.
Let's Talk