Scalable AI Chatbot for a Manufacturer to Solve Customer Support Delays
24/7
Customer support
Cover that no longer depends on someone being available to answer.
70+
Products in the portfolio
The range the assistant has to answer questions about.
20+
Export countries
Markets the client sells into, which is why multilingual handling mattered.
The client
A heating systems producer with a portfolio of more than 70 products and exports to over 20 countries.
Its customer support ran on manual processes, which is where this project started.
The challengeWhat the teams actually needed
Support leaned on manual work, which drove costs up and response times down, while the information people needed sat scattered across systems.
Challenge 01Manual answers, slow and costly
Every response was written by hand, which pushed operational costs up and reaction times down, with nothing covering customers outside working hours.
Challenge 02Answers scattered across systems
Dispersed information made accurate answers hard to reach, and standard systems could not adapt to a user’s context.
Challenge 03No view of what support was doing
Insight into query handling and the effectiveness of support was limited.
What we did
An assistant with a knowledge base the client can edit
The platform runs on an LLM with retrieval-augmented generation, with a dynamic knowledge base and an administration panel that keep content in the client’s hands.
One responsive interface, every state mapped
A responsive full-screen interface for desktop and mobile, with wireframes and mockups covering new chats, resumed ones, error states and animations, and user journeys mapped so navigation stays obvious.
Retrieval wired into the existing CMS
The backend manages sessions and conversations, pulls data from the CMS through agent tool calls, handles multilingual queries and exposes API endpoints to the frontend, with the knowledge base updated straight from the CMS.
A standalone chat app, mobile first
Responsive layouts, conversations updating in real time, and interactive touches such as an “AI is typing…” notice and a history of past chats.
Shipped on its own subdomain
Pipelines, environment setup and automated tests carried each release, with QA spanning functionality, security, performance and user testing on a range of devices and browsers. Going live on a dedicated subdomain helps scalability, branding and SEO.
Tech stackThe tools that made it possible
01
LLM with RAG
The architecture behind the platform, keeping answers tied to the client’s own content.
02
Existing CMS
The dynamic knowledge base runs through the CMS already in place, so updating content needs no developer.
03
CI/CD
Pipelines, environment setup and automated tests behind every deployment.
More like thisOther places we scaled product teams for growth.
What’s next