Dialflo — Building a self-serve voice AI platform from zero
By December our biggest client was about to fire us. The real problem was not in our code. It was in the cab of a truck on a long haul, where drivers could not deal with a clunky scripted call.
Dialflo is a voice AI startup in Bangalore. When I joined, the co-founder had already left. It was the founder, one engineer, and me, with no product and a folder of Python scripts. This is what I built, and how I saved that client.
why I joined
A friend had started a voice company and wanted me as one of the first people in. I came with the zero-to-one experience from Jan Ki Baat, and I wanted to do it again.
The original idea was a LinkedIn-style network for blue-collar workers. That is what pulled me in. It later pivoted to general voice AI. I knew it would take all of my time, and I took it anyway.
what I walked into
A team of three. A LiveKit pipeline existed. Everything else was manual Python scripts. To make a single call you ran a script and set up the campaign by hand. There was no dashboard and no product.
The first work in market was ONEST calls in Kannada and a few Hindi pilots. I already knew the landscape from Jan Ki Baat, where I had worked with more than ten vendors and platforms. We rented an office and built the company from there.
every hat
On paper I was the founding member for Product and Engineering. In practice I worked like a co-founder. I owned the product and customer success end to end. I ran engineering daily, two engineers and a 10am standup.
I also ran the back office. I wrote offer letters, managed the CA, handled GST filing, and built the expense sheets. I visited client offices, recovered churned clients, and presented Dialflo on stage. A 9am founder call and a 10pm founder call, six days a week. No trip in nine months. I still trained, marathons and cyclethons.
what I built
A real self-serve dashboard, so nothing depended on the engineering team to run a call. I started with a Lovable prototype of the look, then built the frontend and the full flow, from signup to agent creation to launching campaigns.
The dashboard does the work: call insights, quality disposition, campaign and agent analytics, account creation, automatic insights. I automated everything that used to be manual, auto-redial on call outcome, number swapping, lead status, launch logic. I added per-call evals so testing was easy, and built agents that improve each week on their own call data. It is live and in use, and it is a real competitor to the existing vendor.
the save
One client was about to churn. The end user runs GPS devices on vehicles inside a geofence. When a truck crosses the fence, an agent calls the driver to get the GPS submitted. Calls were dropping, the script was rigid, conversion was bad, and they wanted around 90 percent. By December the account was nearly gone.
So I went to their office and sat for four hours. I learned the actual business: pickup rate, success rate, the best time to call, how often they switch numbers. I listened to live calls and to the people who had been making them.
The drivers were hanging off the side of the truck on long inter-city hauls, full of questions, and a rigid script could not hold them.
I redesigned the whole conversation flow and the use case, switched to a female voice, and fixed the call quality. I set realistic weekly targets and put their expectations and the reality on the same table. The proof-of-concept worked. A week's trial worked. I went back, showed the win, and got four more projects from the same team. One churning account became five live projects, and we are now automating their control tower and GPS operations end to end, live in their system.
direction
It is a crowded market, so I moved us off flat pricing to outcome-based pricing, then to implementation fees, and pointed the company at transportation and hiring. From January we rebuilt from scratch and went from zero to four paying clients. The company was older than that, but on a short runway. I treated it like a zero-to-one build anyway.
what I took from it
I ran a company, not a feature. Product, customer success, engineering, operations, the books, all of it at once. This was the second best thing I have built, after Jan Ki Baat. The save is the lesson I keep: you do not fix a churning account from a dashboard, you fix it by sitting in their office for four hours and learning the real business.