A real founder story about how Elba started as a healthcare voice AI product, hit the wall of slow medical procurement, and was rebuilt into a configurable platform for regulated industries. Told through Slack moments from the team’s pivot, the post shows how healthcare’s hardest constraints became Elba’s moat across insurance, finance, public services, and emergency assistance.
Virendra Singh Bhalothia
8 min read
May 19, 2026
About the Author
Virendra Singh Bhalothia
Engineering Reliability & Innovation | It's time to build.
He was at Latitude59 and had just spent two hours with a VC partner he'd clicked with. The conversation had nothing to do with raising money. The guy was telling Ben how something like Elba would change the way he ran his week, and asking, half-curious, whether we'd ever build it for anyone outside healthcare.
Ben in Tallinn
Ben told me he wasn't the only one asking. All week in Estonia, people in insurance, finance, and law kept circling back with the same question. Nobody was building voice AI with the compliance and sovereignty posture we'd built. They weren't asking for features. They were asking us to exist for them.
I remember the call being careful. Something like, I think we have a problem. Or an opportunity. I'm not sure yet.
Three weeks later he was at Viva Tech in Paris and the signal was louder. Insurance. Banks. Law firms. Roadside assistance. The call I got this time wasn't careful. It was a decision already made.
We can't wait. German healthcare procurement will take two years and we don't have two years. Find out what it takes to open Elba up.
That's where this story actually starts. Not in a Slack thread. Not in a strategy doc. On the phone, with one of us in another country, telling the other one that the company we were building was not the company we actually were.
Why we built for healthcare in the first place
Ben and I had both been living in Hamburg for seven years when we started Kolsetu. Neither of us moved here to build a company. We just lived here, did our work, built lives, and at some point realised that the things that quietly broke for us as foreigners also quietly broke for millions of other people.
For Ben, the breaking point was a stretch of bad health and a long encounter with the German medical system. Trying to navigate that system without German is a particular kind of helplessness. He came out the other side knowing exactly what he wanted to build.
We knew from the first whiteboard that, as an Israeli founder and an Indian founder, we'd need to be 500% more secure and compliant than anyone needed us to be. Germans get the benefit of the doubt. Foreigners don't. If we wanted to be in the room, we had to walk in already over-prepared.
So we built Elba for German medical clinics. FHIR-aligned data models. ISO 27001 before revenue. EU-sovereign hosting. The kind of infrastructure that takes a year of work and gives you nothing flashy to show.
The product worked. Clinics loved it. Patients loved it.
One problem. Healthcare couldn't buy us at the speed we needed to survive.
The wall
German clinics operate at 120% capacity. Procurement is not on their list of things they can do this quarter, no matter how badly they want what you're selling. We'd demo, they'd love it, then nothing for months.
Meanwhile the people Ben met in Tallinn and Paris kept calling. Insurance companies. Roadside operators. Banks. They'd heard about our compliance posture and wanted the same thing for their conversations. They didn't want a pilot. They wanted contracts.
Ben's read on it was sharp and unsentimental. We were a bootstrapped startup with a survival problem. The market was telling us where the oxygen was. The only acceptable response was to go where the oxygen was.
There wasn't a long debate about it. There wasn't an identity crisis on his side. He called from Paris in war-time founder mode, the way founders sound when they've already decided and are now telling you what comes next. Rip the bandage off. Open Elba up. Keep pushing on healthcare in parallel because that's the core mission, but don't let it sink the boat.
The identity crisis was mine.
Six weeks, thirty thousand lines
When Ben called from Paris, Arpit and I sat down with what we had and what we needed and started to count the gap. Elba had been built as a healthcare product, end to end. The data models were medical. The table names were medical. The prompts were hardcoded for clinic workflows. Every architectural decision assumed one kind of customer.
What Ben needed was a platform that could run a medical receptionist, a legal intake agent, an insurance overflow line, and an emergency dispatcher from the same core, with the same compliance guarantees, configured differently per domain.
We had six weeks.
Hackathon in India.png
This was the early days of Claude Code and Cursor. Not the version people are using now. The tooling was rough. The mistakes were expensive. We were not riding a wave of AI productivity. We were rebuilding a production system live, with paying healthcare customers on the existing version, while reshaping the architecture underneath them.
Thirty thousand lines of code over six weeks. Arpit drafted a thirteen-step implementation plan that we ran like a sprint inside a sprint. We ripped out healthcare-specific data types and replaced them with generic structures. Built a workflow engine driven by JSON configuration files. Created a node-based execution system where the behaviour of the agent comes from config, not code. Replaced hardcoded prompts with templates selectable by domain.
One of our pull requests from that period had a six-word description that I still think about: Fixed the medical related types to generic. Six words. Days of work. An entire identity rewritten in present tense.
The pattern we finally saw, the one our codebase had been screaming at us for months, was this: every customer we'd ever talked to - medical, legal, financial, insurance - needed 80% the same infrastructure and 20% domain-specific configuration. Our old code had treated everything as 100% healthcare. Every custom feature request had been a configuration trying to escape the codebase.
We hadn't been building healthcare software. We'd been building trust infrastructure. Healthcare was the first configuration of it.
The technical work was hard. It cost sleep and weekends and the kind of focus that wrecks your appetite. But the harder part for me, sitting at the keyboard with Arpit through those weeks, was watching the company I thought we were dissolve and reshape in real time. Ben was already on the other side of that. I had to catch up.
What we did not let go of
Ben drew one line and held it. We were not abandoning healthcare. We were refusing to let healthcare kill us before we could come back to it.
He kept pushing on hospital conversations. He kept showing up at health networks. He kept writing to clinics. He still does. Every week there's another healthcare meeting in his calendar that none of the rest of us would have made happen. The mission didn't change. The strategy did.
What we stopped doing was staking the entire company on a single buyer who needed two years to sign.
What happened next
A Vienna Insurance Group subsidiary, Global Assistance, signed with us in central Europe. They run roadside and emergency assistance, 300,000+ cases a year, 700+ rescue vehicles. The kind of operation where a missed call is a person stranded on a motorway.
We deployed Elba to take their overflow. When every one of their human reps is on another call, the next caller reaches Elba instead of a hold tone. Of the calls Elba takes, 85% are resolved end-to-end with no human handoff. We're tracking toward 90%. 100% uptime since go-live. Six to nine weeks from contract to production. 100+ languages.
The architecture Ben asked Arpit and me to build on a phone call from Paris is now answering roadside emergencies across central Europe in the middle of the night.
What I'd tell another technical co-founder
The market you build for is not always the market that buys from you. The signal is in who calls you back fastest.
Build for the hardest customer first. The constraints they put on you become your moat the moment you walk into any easier room.
When every feature request looks like a configuration change trying to escape the codebase, your architecture is telling you what you actually are. Listen to it earlier than I did.
If your CEO calls you from another country and tells you it's time to rip the bandage off, you don't get to keep the bandage.
We built the wrong company right. Healthcare wasn't a mistake. It was our proving ground. And we're still going back to it, one clinic at a time, with a platform that can finally survive the wait.
Virendra Singh Bhalothia
