Advanced AI Agent for a roofing contractor — self-learning RAG, sales consulting, quote closing, survey scheduling, proposed carts, bills, and full checkout. Trainable, self-correcting, and operator-ready.
Project Overview
AI Agent & Sales Bot is a powerful, production-grade AI Agent for DEVO Roofing & General Contractors — not a FAQ widget, but a sales and operations assistant that consults shoppers, closes quotes, schedules roof surveys, proposes carts, builds bill lines, and completes checkout. Private deployment — no public demo URL.
Role & Delivery
End-to-end AI product: hybrid RAG corpus, curated teachings, feedback-driven learning, multilingual shop assistant (EN / VI / ES), admin operator console, and Stripe-backed commerce hooks.
Core Capabilities
- Self-learning RAG — Products, FAQs, crawled pages, and teachings indexed for hybrid retrieval; vectors embed automatically
- Interactive consulting — Shingles, tools, Texas delivery, member pricing, coupons
- Sales closing — Guided turns that move toward purchase or a paid quote
- Survey appointment planning — Schedule on-site roof surveys; quote & survey leads in admin
- Autonomous sales advising — Intent handling and recommendations across locales
- Bills, cart & checkout — Proposed cart cards in chat, line edits, confirm → cart, full checkout
- Self-correction — Thumbs-down queue → curated fix → promote to AI Training
- Trainable by design — Operators add teachings, import glossaries, learn from URLs, and crawl knowledge sources
Technology Stack
| Layer | Choices |
|---|
| App | Next.js, React, TypeScript, Tailwind, shadcn/ui |
| AI | Shop assistant LLM, hybrid RAG, embeddings, reply critic |
| Ops | AI Training, knowledge sources, quality dashboard, feedback queue, chat logs |
| Commerce | Proposed cart → cart → Stripe checkout; quote & survey leads |
Admin AI Screenshots
Operator console under SEO & AI — every settings tab plus training, knowledge, quality, feedback, and logs.











Outcome
A modern AI sales agent for a roofing brand — trainable, self-improving, and wired into real commerce and survey workflows. Private deployment — no public demo URL.