Most companies today need an intelligent chatbot on their website for customer support, FAQs, and lead handling. However, building a chatbot from scratch requires AI expertise, backend infrastructure, vector databases, document processing, and continuous maintenance.
To solve this, I built a plug-and-play AI chatbot platform where companies can create, train, and deploy a chatbot without building one themselves.
Users simply:
- Upload documents or
- Add website links and paste a single script — the chatbot becomes live instantly.
The chatbot uses Retrieval-Augmented Generation (RAG), ensuring accurate, document-based answers instead of generic AI responses.
- Provide a ready-to-use AI chatbot platform
- Enable RAG-based question answering
- Allow document + website trained chatbots
- Easy embed & deploy on any website
- Reduce AI & development complexity for companies
This platform is built using a Retrieval-Augmented Generation (RAG) approach instead of plain OpenAI chat.
-
Documents & website content are:
- Extracted
- Chunked
- Converted into vector embeddings
-
User question is:
- Embedded
- Compared with stored vectors
-
Most relevant chunks are retrieved
-
Final answer is generated only from retrieved content
-
Prevents hallucination
-
Answers are grounded in user-provided data
-
Supports multi-source knowledge (documents + websites)
- Backend: Node.js, Express.js
- Database: MongoDB (Documents, Chunks, Chat History, Tickets)
- Vector Storage: MongoDB Vector Embeddings
- Frontend: React.js (Dashboard + Widget)
- AI Engine: RAG (Embeddings + Context Retrieval)
- Authentication: JWT Authentication
- Security: Middleware-protected APIs
- JWT-based authentication
- Email verification during signup
- Protected APIs for:
- Chatbot queries
- Document uploads
- Website scraping
- Chat history
- Tickets & dashboard data
The platform provides a complete Admin/User Dashboard with the following modules:
- Store user profile information
- Manage chatbot-enabled websites
- Each user has isolated chatbot knowledge
- Upload
.docxdocuments - Documents are:
- Extracted
- Chunked
- Embedded
- Stored in vector format
- Chatbot answers questions strictly based on uploaded documents
- Supports:
- single document
- all-document queries
- Add website URLs (e.g. documentation pages)
- Automatically:
- Crawl pages
- Extract clean text
- Split into chunks
- Generate embeddings
- Website content becomes chatbot knowledge
- Manage:
- View added links
- Delete scraped links
- Answers are generated using:
- Retrieved document chunks
- Website content
- No random or hallucinated responses
- Supports:
- Context-aware conversations
- Multi-document understanding
- Stores all chatbot conversations
- Analyze:
- Total number of questions asked
- Most common questions
- User interaction trends
- Useful for improving chatbot quality
- Users can raise support tickets directly from chatbot
- Admin can view:
- User name
- Issue description
- Ticket status (Pending / Done)
- Admin can mark tickets as Done
- Auto-generate chatbot embed script
- Users just need to paste the script into their website
- Works with:
- HTML projects
- React apps
- Any MERN or static website
- Floating chatbot widget (bottom-right corner)
- Features:
- AI Chat
- Ticket Raise option
- Small conversation UI
- Easily customizable
- Can be tested locally or on production websites
- Test chatbot using:
- Local HTML file
- React frontend
- Works with:
- Localhost
- Production servers
- Easy integration without backend changes
- Backend architecture completed
- RAG pipeline implemented
- Document upload & embedding working
- Website scraping & crawling working
- Chatbot APIs ready
- Dashboard UI improvements - in progress
- Widget UI enhancements - in progress
- FAQ auto-suggestions
- Multi-language chatbot
- Role-based admin access
- Advanced analytics dashboard
- Live agent support integration
- Vector database optimization