Verto AI is a modern AI-powered customer support platform that enables businesses to deploy an intelligent chatbot trained on their own knowledge base, manage customer conversations, and escalate complex queries to human agents.
Built as a reusable SaaS starter kit for AI customer support systems.
🚩 Problem
Customer support teams face:
- High volume of repetitive questions
- Slow response times
- High operational costs
- Limited automation
Most businesses want AI chatbots, but existing solutions are either expensive, hard to customize, or lack deep knowledge-base integration.
🎯 Goal
Build a platform where organizations can:
- Upload their documents
- Instantly deploy an AI chatbot
- Answer customer questions accurately
- Escalate complex cases to humans
- Manage everything from a dashboard
✅ Solution
I designed and built a multi-tenant AI-powered SaaS platform that provides:
- AI chatbot using Google Gemini
- RAG-based knowledge retrieval
- Embeddable website chat widget
- Admin dashboard for conversation management
🧱 Core Features
AI-Powered Chat
- Google Gemini 2.5 powered responses
- Context-aware conversations
- Automatic human escalation
Knowledge Base
- File upload
- Document chunking
- Embeddings storage
- Semantic search
Customer Widget
- Embeddable JavaScript snippet
- Custom greetings & suggestions
- Persistent sessions
- Responsive UI
Admin Dashboard
- Organization-based multi-tenancy
- View / resolve / escalate conversations
- Widget customization
- File management
🏗 Architecture
Monorepo using PNPM workspaces and Turborepo.
Applications
apps/web→ Admin Dashboardapps/widget→ Chat Widgetapps/embed→ Script Loader
Packages
packages/backend→ Convex schema & AI logicpackages/ui→ Shared components
🔧 Tech Stack
Frontend
- Next.js 15
- Tailwind CSS
- Shadcn UI
- React Hook Form + Zod
- Jotai
Backend
- Convex (DB + serverless)
- Convex RAG
- Convex Agent
AI
- Google Gemini 2.5
Auth
- Clerk
Monitoring
- Sentry
🔁 Data Flow
- Admin uploads documents
- Documents are chunked & embedded
- User sends message from widget
- Query searches embeddings
- Context injected into prompt
- Gemini generates answer
- Optional human escalation
🧩 Database Design
- organizations
- users
- conversations
- messages
- knowledge_documents
- embeddings
- widget_config
Each entity is scoped by organization.
⚙️ Key Technical Challenges
Multi-Tenancy
Each organization must have isolated data and configuration.
Solution:
All tables scoped by organizationId.
Real-Time Messaging
Need instant updates between widget and dashboard.
Solution:
Convex real-time subscriptions.
Accurate AI Responses
LLMs hallucinate without context.
Solution:
RAG pipeline with document embeddings.
Embeddable Widget
Should work on any site.
Solution:
Script loader that injects iframe-based widget.
📈 Result
- Fully functional SaaS MVP
- Production-ready architecture
- Reusable starter kit for client projects
💡 Learnings
- Designing AI systems requires strong data pipelines
- RAG dramatically improves accuracy
- Monorepo architecture improves reuse
🔮 Future Improvements
- Sentiment analysis
- Multilingual support
- Voice messages
- CRM integrations
🔗 Links
� Need Something Similar?
Looking for an AI-powered customer support solution or a similar SaaS platform for your business? I'd love to help you build it.