Voice to professional estimate in seconds. Built for contractors in the field.
Hackyard 2026 — Built live August 28–30, 2026.
Contractors lose jobs because estimates take too long. A framer finishes a walkthrough, drives home, opens Excel, and spends an hour typing. By then the homeowner already called someone else.
Open the app. Hit record. Describe the job out loud — materials, labor, square footage. Done. A professional PDF estimate is generated in seconds, ready to email before you leave the driveway.
| Layer | Technology |
|---|---|
| Backend | Python / Flask |
| Speech-to-Text | faster-whisper (Whisper small, GPU) |
| AI Parse | Qwen3 8B via Ollama (local, zero cost) |
| PDF Generation | ReportLab |
| History | SQLite |
| Gmail SMTP |
- 🎙 Voice input — record your job description on any device
- ⚡ Instant parse — Qwen3 extracts line items, quantities, units, and prices
- 📄 PDF generation — professional contractor estimate, download instantly
- 📧 Email to customer — send the PDF before leaving the job site
- 📋 Estimate history — every estimate saved, retrievable anytime
- 💰 Customizable rates — set your own labor rates per trade
- Python 3.10+
- CUDA-capable GPU (recommended) or CPU
- Ollama running locally with
qwen3:8bpulled - ffmpeg installed
git clone https://github.com/Jabergan/voiceestimate.git
cd voiceestimate
pip install -r requirements.txt
ollama pull qwen3:8bcp .env.example .env
# Edit .env and add your Gmail app passwordbash start.sh
# App runs on http://localhost:5055- Open the app in your browser
- Enter contractor name, customer name, and job address
- Hit the microphone button and describe the job
- Hit Parse Line Items
- Review and adjust if needed
- Hit Download PDF or Email to Customer
No API costs. No latency. No data leaving the job site. Qwen3 8B runs entirely on local hardware — a $400 GPU handles it in under 3 seconds.
Built by Jamie Bergan — Minot, ND