-
-
Operations dashboard summarizing quote volume, automation, readiness, risk distribution, and follow-up activity.
-
Manual request capture for calls, text messages, walk-ins, forms, copied email, and uploaded request documents.
-
Sent-reply review with delivery state, readiness evidence, risk gate, and follow-up controls.
-
Configurable sync schedules, platform scope, reply sending modes, and safety controls for autopilot behavior.
-
Live Alibaba Cloud deployment showing QuotePilot prioritizing unanswered customer requests and separating already answered messages.
-
Business profile and price book grounding used by Qwen to produce accurate, policy-aware quotes.
-
Inbox triage view with service-request filtering, search, sorting, deletion controls, and workflow conversion.
Inspiration
Small service businesses often lose time and leads because customer requests arrive from everywhere: email, forms, text messages, calls, and copied notes. A simple question like “Can you fix my sink today and what will it cost?” can turn into several manual steps: understand the job, check the price book, ask for missing details, draft a reply, get approval, send the message, and remember what happened.
QuotePilot was built for Track 4: Autopilot Agent to show how an AI agent can complete that workflow end to end, while still keeping the business in control.
What it does
QuotePilot turns inbound customer messages into quote workflows.
It can:
- Sync incoming customer requests from connected channels.
- Filter out non-service messages without wasting AI credits.
- Triage real service requests and identify the job type, urgency, missing details, and customer intent.
- Use a business profile and price book so quotes are grounded in the business’s real services and rates.
- Draft customer replies with evidence, memory updates, and delivery status.
- Support manual approval, guarded auto-send, or always auto-send modes.
- Keep a review trail of sent replies so the owner can see what happened.
- Let businesses configure schedules, channels, contact rules, holidays, active dates, pause windows, and platform scope.
The goal is not just to draft text. The goal is to move from customer message to approved quote with practical automation that a real small business could use.
How we built it
QuotePilot is a Node.js web app with a custom frontend, local data store, email connector, webhook-style adapter support, and Qwen-powered reasoning through Alibaba Cloud Model Studio.
Qwen is used for the parts that need understanding: extracting the customer’s request, classifying service intent, identifying missing information, grounding the quote against the price book, and drafting a safe reply. Deterministic rules handle cheaper operational work such as spam protection, duplicate detection, schedule checks, restore lists, and safety gates so API credits are used responsibly.
The app includes:
- Qwen 3.7 Plus integration through the OpenAI-compatible API.
- A business setup area for importing profiles and price books.
- Quick price rows for businesses that want spreadsheet-like setup.
- Inbox protection and restore tools.
- Multi-schedule automation with platform scope, contact targeting, date rules, public holiday handling, and override behavior.
- Reply sending modes for manual, guarded, and fully automatic workflows.
- A workflow review screen for quote evidence and sent-reply audit.
Challenges
The hardest part was making the product feel practical instead of like a demo. Real businesses do not want a chatbot that invents prices or sends risky replies. They need control, traceability, and simple setup.
Some of the biggest design challenges were:
- Preventing the app from reprocessing the same messages repeatedly.
- Avoiding unnecessary Qwen calls during refreshes.
- Making auto-send powerful but still safe.
- Supporting different future message platforms without hardcoding the product around one provider.
- Letting schedules handle real business complexity without overwhelming the UI.
- Keeping price-book grounding clear enough for users and judges to trust.
What we learned
A useful autopilot agent needs more than a model call. It needs memory, rules, tools, schedules, permissions, and a clear human override path. Qwen is strongest when it is given clean business context and used for reasoning, while deterministic logic handles repeatable operational checks.
We also learned that trust is a product feature. QuotePilot shows why it made a decision, what quote it drafted, whether it was sent, and what information is still missing.
What’s next
Next we would expand the connector system so businesses can plug in more channels such as website forms, CRMs, support inboxes, SMS providers, and messaging platforms. We would also add hosted multi-tenant accounts, richer analytics, team permissions, and deeper integrations with calendars, invoices, and payment tools.
QuotePilot’s long-term vision is a small-business operations desk where routine customer quoting runs automatically, but the business owner always knows what happened and why.
Log in or sign up for Devpost to join the conversation.