LeadDesk AI is an independently built Noerong product for service businesses that repeatedly answer the same questions and receive incomplete quote requests. I designed and developed the customer assistant, enquiry workflow, private dashboard and deployment package. The demonstration uses fictional BrightHome Cleaning data.
The problem
A visitor needs an answer before asking for a quote. The business needs enough context to follow up. I connected those two steps without allowing the assistant to invent prices, policies or services.
What I built
A branded website assistant and embeddable widget.
Approved FAQ retrieval with source links and a fallback for unknown questions.
A validated quote form, lead scoring, status tracking and CSV export.
A token-protected admin dashboard and persistent SQLite storage.
Optional server-side OpenAI responses and webhook delivery to a separately configured destination.
The workflow
Visitor question → approved answer → structured quote request → validation → private lead dashboard → follow-up or export.
Decisions that matter
The assistant works without a paid AI provider. Business facts come from approved content, credentials stay server-side, and AI response storage is disabled by default. The deployment is scoped to one small business; team roles and multi-tenant operation require additional engineering.
Evidence and current scope
Thirteen automated tests passed across the application, retrieval, AI fallback and deployment configuration. This demonstrates the tested workflow, not a measured increase in customer sales. Python, FastAPI, SQLite and Docker support a buyer-owned deployment with persistent storage.