Work
Things we have built and still run.
Much of our client work sits under NDA. What follows is what we can show — starting with a product we own end to end, which means we can be specific about how it is built.
Product · Built and operated by IR4
Onramp
A conversational AI that answers enquiries, qualifies them and books work for service businesses — over WhatsApp, because that is where their customers already are.
The problem
Service businesses lose work in the gap between an enquiry arriving and somebody being free to answer it. The enquiry comes in during a job; by the evening the customer has booked with whoever replied first. Answering services are expensive and cannot qualify the work or see the diary.
What we built
A system that holds a real conversation on WhatsApp: understands what the customer is asking for, asks the questions a competent receptionist would ask, checks availability and books the job. It escalates to a human when it should, rather than guessing — which in practice is the difference between a useful assistant and an expensive liability.
What changes for the business
Stripped of the technical detail, this is what the people using it actually notice:
- Enquiries get answered while the team is on a job, instead of that evening or the next morning.
- The questions that decide whether a job is worth quoting get asked every time, rather than when someone remembers.
- Bookings land in the diary directly, so nobody transcribes them from a phone at the end of the day.
- Anything unusual, awkward or worth a human conversation is passed to a person with the context attached.
- Nobody spends their evening replying to messages that could have been handled at two in the afternoon.
None of that is a claim about a percentage. It is a description of what the system does — the value of it depends entirely on how many enquiries a business is currently losing in that gap, which is a question worth answering before buying anything.
Why it is worth showing you
Anyone can demonstrate an LLM answering a message. Onramp is the other 90%: conversation state that survives a customer replying three hours later, idempotent booking so a duplicate message cannot double-book, monitoring that tells us before a client notices, and a cost model that works at real message volumes. It runs on IR4’s own Kubernetes infrastructure.
Visit getonramp.techBuilt with
- TypeScript
- Node.js
- PostgreSQL
- Redis
- Kubernetes
- WhatsApp Business API
- LLM orchestration
Role
Design, build, operate
Status
Live in production
Hosting
IR4 self-hosted Kubernetes
Look under the bonnet
Because we own this one outright, there is nothing we can’t show you. We will walk you through the architecture, the database design, the monitoring, what it costs to run and the decisions we got wrong the first time — in whatever technical depth you or your own developers want.
That is a harder test than a case study, and a more useful one.
Book a technical walkthroughBackground
Where the engineering comes from.
High-value systems
A decade at The Hut Group, including technical leadership on a £100M greenfield platform — taken from first commit through go-live and into long-term operation.
Regulated and secure
Earlier work in insurance software under PCI-DSS, including a customer file-upload system that stood up to four years of professional penetration testing.
Scale and resilience
Distributed, fault-tolerant, message-driven architectures — event sourcing, CQRS, high availability. The habits that come from systems where downtime is measured in money.
Want to talk about something similar?
We are happy to walk through how any of this was built, in as much technical detail as you want.
Start a conversation