Multi-channel Chatbot
AI-powered chatbot with RAG (retrieval from your real internal knowledge base), multi-channel support, and Vietnamese NLU. Designed for SMEs: easy to operate, no technical team required. We build only the features your business actually needs, aligned to your real business model, nothing more, nothing less.
Common problems
Not enough staff to handle customers 24/7, especially outside business hours
Internal documentation is extensive and constantly changing, so new hires take weeks to get up to speed
Customers repeatedly ask simple questions that require accurate answers based on the latest policies
Losing leads due to slow responses: customers switch to competitors before your team can reply
How MADIAD solves it
How one question travels through the system
Eight layers, run in order. The question walks down, the answer turns around and comes back on the channel the customer is already using.
- A customer sends a message
- 01Interfaces and channels
The bot answers wherever the customer wrote, so nobody has to switch channels. It accepts text, voice, images, video and file attachments.
- Website and app
- Zalo OA and personal Zalo
- Telegram
- Voice: speech-to-text and text-to-speech
- 02Security gateway and controls
Every message passes one door. That door authenticates the sender, throttles flooding and filters harmful content before the model ever reads it.
- API gateway
- Rate limiting
- OAuth2 / JWT auth
- Prompt-injection filtering
- 03Orchestration and agent layer
Reads the real intent behind the wording, breaks a large request into steps, routes each to the right agent and keeps the state of the conversation.
- Intent recognition
- Multi-step planning
- Multi-agent coordination
- State management
- 04Memory
Nobody has to start over. The bot holds the running conversation, remembers a customer profile across visits, and recalls by meaning rather than by keyword alone.
- Session memory
- Long-term customer profile
- Semantic memory
- 05Knowledge retrieval (RAG)
Answers come from your own documents: price lists, policies, product descriptions, procedures. Retrieval combines keyword and semantic search, then reorders results so relevant passages come first.
- Vector database
- Hybrid search
- Reranking
- Source citations
- 06Tools and external APIs
The bot does more than talk: it looks up orders, books appointments, records leads, runs calculations. Each action is a controlled call into your systems.
- Function calling
- ERP and CRM connectors
- Web lookup
- Code interpreter
- 07Language model
Where the answer is written. The model reads several kinds of input, and can run on your own infrastructure when your data is not allowed to leave the building.
- Multi-modal
- Self-hosted model option
- Domain fine-tuning
- 08Post-processing before sending
The answer is reviewed before it reaches the customer: checked against sources, personal data masked, out-of-authority statements blocked. Wrong answers get rewritten, out-of-scope ones go to a person.
- Grounding check
- PII masking
- Output guardrails
- Self-correct or hand off
- The answer returns on the channel the customer is already using
The diagram describes what each layer does. The specific technology behind a layer is chosen against your requirements, your data and the infrastructure you already run, and is agreed during the survey.
What the bot can do
Not a simple chatbot. This is a complete AI system.
Retrieves information from your business's own internal knowledge base: products, pricing, policies, FAQs. No hallucination, always answers based on real data.
Understands context, punctuation, and Vietnamese abbreviations. Distinguishes intents: pricing inquiry, booking, complaints, feedback.
Connects directly to CRM, Google Calendar, Google Sheets, email systems. The chatbot can create bookings, update leads, and send confirmation emails.
Not simple if-else. A decision tree system combined with AI: evaluates intent → selects flow → executes → verifies results.
Knows when to transfer to a human: negative sentiment, out-of-scope questions, complex requests → notifies team + transfers full context.
One system, many channels: Facebook Messenger, Zalo OA, Telegram, WhatsApp, website widget. Centralized management, synced conversation history.
What you get
- Multi-channel chatbot: Website, Facebook, Instagram, TikTok, LinkedIn, Telegram, Zalo personal, Zalo OA *
- Monitoring dashboard + conversation analytics
- Custom training dataset based on your business knowledge base
- Documentation + handover training for your operations team
* Some platforms require an approved business account with API access. MADIAD handles the entire registration, verification, and integration process.
What we need from you
- Product and service information to load into the knowledge base
- Brand voice and how you want customers addressed
- The channels you want live: website, Zalo, Messenger
- For Messenger: a Facebook Page and admin access
- For Zalo: an account we can sign the session in with
What to know before you start
- Messenger blocks outbound messages more than 24 hours after the customer's last message — that is Meta policy, not a limitation of the system. Past that window, someone reaches out by hand.
- Web chat has no outbound path: once the visitor closes the tab the thread is gone, unless they left a phone number or email.
- Zalo runs on a personal login session, not the Zalo OA API. Sessions expire and have to be re-scanned; a heartbeat watches for a dead session so it is caught early.
- The bot does not quote prices, agree prices, or promise deadlines. It captures what you need and hands you to a person — by design.
- The bot says it is an AI. It does not pose as a human.
Multi-channel Chatbot
AI-powered chatbot with RAG (retrieval from your real internal knowledge base), multi-channel support, and Vietnamese NLU. Designed for SMEs: easy to operate, no technical team required. We build only the features your business actually needs, aligned to your real business model, nothing more, nothing less.