Personalised Conversations at Scale
How RAG and Model Context Protocol turned a generic chatbot into a trusted team member that understands context, connects to business systems, and delivers answers people actually trust.
Generic AI was not good enough
PingMe's customers were already using AI chat tools, but the experience was frustrating. Off-the-shelf chatbots gave confident-sounding answers that were often wrong. They could not access real customer data, had no awareness of business context, and treated every conversation as if it were the first.
Support teams spent as much time correcting the AI as they saved by using it. Customers lost trust. The promise of AI-powered support was undermined by hallucinations, stale information, and an inability to actually do anything beyond generating text.
Ground every answer in real data, then give it hands
We built PingMe's AI assistant on two foundational ideas. First, retrieval-augmented generation (RAG) to ensure every response is grounded in actual business data, not the model's training data. Second, Model Context Protocol (MCP) to give the assistant the ability to read from and write to the systems the business already uses.
The result is an AI that does not just talk - it acts. It looks up orders, updates addresses, books appointments, and creates tickets. And it does all of this across WhatsApp, SMS, email, and web chat with a consistent voice and context that carries across channels.
One brain, four channels
The same AI, the same knowledge, the same capabilities - regardless of how the customer reaches out. Conversation context persists across channels, so a customer who starts on WhatsApp can follow up via email without repeating themselves.
End-to-end encrypted conversations with rich media support, quick replies, and template messaging via the WhatsApp Business API.
SMS
Twilio-powered SMS with intelligent message chunking, delivery receipts, and opt-in/opt-out compliance built in.
Threaded email conversations with HTML formatting, attachment handling, and automatic ticket creation for complex queries.
Web Chat
Embeddable chat widget with typing indicators, conversation history, and seamless handoff to human agents when needed.
How every answer stays grounded
Retrieval-augmented generation connects the language model to real business data at query time. Instead of relying on what the model memorised during training, it retrieves the most relevant information from the company's own knowledge base before generating a response.
- PDFs, knowledge bases, and help docs
- Product catalogues and pricing data
- CRM records and customer history
- Internal wikis and SOPs
- Chunking with overlap for context preservation
- Embedding generation via sentence transformers
- Metadata extraction for filtered retrieval
- Incremental re-indexing on content changes
- Hybrid search combining vector similarity and keyword matching
- Re-ranking with cross-encoder models for precision
- Contextual filtering by customer segment and query type
- Citation tracking back to source documents
- Grounded responses with source attribution
- Tone and brand voice consistency
- Hallucination detection and confidence scoring
- Graceful fallback to human handoff
Not just answers - actions
MCP gives the AI assistant a structured way to interact with business tools and APIs. Instead of just generating text, it can look up data, trigger workflows, and update records - all through a standardised protocol that keeps interactions safe and auditable.
Order Lookup
Query order status, tracking information, and delivery estimates directly from the commerce platform.
Account Management
View and update customer profiles, subscription status, and billing information without leaving the conversation.
Scheduling
Book, reschedule, or cancel appointments by reading and writing to the calendar system in real time.
Knowledge Search
Deep search across internal documentation, FAQs, and troubleshooting guides with relevance ranking.
Ticket Creation
Automatically create, categorise, and route support tickets when the AI identifies issues needing human attention.
Analytics Events
Log conversation outcomes, resolution rates, and customer satisfaction signals for continuous improvement.
AI became a trusted team member, not just a chatbot
The difference between a chatbot and an AI assistant is trust. PingMe's customers now get answers grounded in real data, with citations they can verify. When the assistant takes an action - updating an address, booking an appointment - the customer sees it happen in real time.
Support costs dropped because the AI resolves the majority of queries without human involvement. But more importantly, customer satisfaction improved. People are not just tolerating the AI - they prefer it for straightforward queries because the responses are faster, more accurate, and available around the clock.
The MCP integration means the system grows with the business. When PingMe adds a new tool or data source, a new MCP server connects it to the assistant without retraining the model or rewriting conversation logic.
Queries resolved without human involvement
Average response time across all channels
Customer satisfaction rating
Reduction in support costs
Built on open standards, not lock-in
"Our customers stopped asking to speak to a human. Not because we made it harder, but because the AI actually understood their problem and could fix it on the spot. That is the difference between a chatbot and an assistant."
Want AI that actually understands your business?
We build AI assistants grounded in your data, connected to your systems, and designed to earn customer trust.