Abdullo Khidoyatov and Azizbek Bakhodirov from Uzbekistan spent several years building the outsourcing company Udevs before launching Ucode — a platform that helps developers build and modify application backends using AI. They invested around $300,000 in the project from Udevs’ profits. Today, the service has around 30 paying corporate clients and generates $5,000 in MRR.
For a special joint project by Digital Business and Astana Hub, “100 Startup Stories from Central Eurasia,” Abdullo Khidoyatov explained why he hesitated for several years before bringing the product to the mass market and how a trip to Silicon Valley changed things. We also talked about the team’s expansion into the US and its plans to scale in India and Pakistan.
“To replace just one component, we would have had to throw a year and a half of work in the trash”
– Abdullo, what were you doing before you launched the startup?
– I worked as a developer on different projects for around eight years. For example, I spent five years at MyTaxi, an online taxi-booking service in Tashkent, where I worked my way up from developer to team lead, then CTO, and eventually CEO.
In 2020, Azizbek Bakhodirov and I launched the outsourcing company Udevs. We started taking on website and app development projects. By around our hundredth project, we realized that a huge amount of our time was no longer going into development itself, but into maintaining finished systems, with constant requests to change or add something.
– How did the idea for Ucode come about?
– The idea came after we more or less messed up two projects. One of our clients was Sofia, a large confectionery chain in Tashkent. Today, the company has around 150 branches and more than 4,000 employees. In 2023, they asked us to build an HRMS, a system to automate HR processes across the entire chain. We spent around a year and a half developing it. During that time, the business changed: employees needed to be assigned to specific branches rather than to the company as a whole. To change such a small detail, we would have had to throw a year and a half of work in the trash and rewrite the code from scratch. In the end, we had to abandon the project.
The second failure happened when we were developing a system for several branches of a local clinic. It turned out that the doctors found it inconvenient to use. Redesigning the interface took too much time. The client didn’t want to pay extra, and we couldn’t keep reworking it for free.
It became obvious that when you work on large projects, the hardest and most painful part for developers is making architectural changes, meaning changing the logic and structure of a system that’s already been built. That’s when we decided to launch Ucode.
– How does Ucode work, and what exactly can users do with it?
– We originally built Ucode for ourselves. The idea was to avoid building the backend of every new website or app from scratch and then spending a lot of time reworking it later.
Ucode already includes the core components developers would normally have to set up separately, such as authentication, access permissions, databases, APIs, backend functions, and other tools. You can put them together and configure them for a specific project, then make changes whenever needed.
– When did you start developing Ucode, and how long did it take to get a working version?
– We started working on it around 2024. Six months later, we had an MVP. We gave all our friends and acquaintances access to the site so they could test it out. That’s when the bugs started showing up.
The system could process a request incorrectly and return a “500 error” to the user. It also struggled under heavy load. Then we ran into another issue: the interface. As a developer, I understood where to add a table or a new field, but employees at the companies that later used the system we had built, such as finance staff or administrators, found the interface difficult to navigate. So we had to redesign the UX separately.
We gradually fixed the issues, and after about a year, Ucode became stable enough to sell.
“People don’t need overloaded functionality. They need a ready-made solution to their problem”
– Who were the first users of Ucode?
– At first, we used the platform ourselves. Since the beginning of 2025, we’ve built around five Udevs projects with Ucode. For example, we put together one CRM system in about two weeks. At the time, a full-scale launch still felt like a distant dream, something we had a long way to go before reaching.
– Why did you feel that way?
– I felt the product still wasn’t good enough. I constantly wanted to add new features and improve the existing ones. That continued until September 2025, when I went to the US as part of the Silicon Valley Residency program, organized by Astana Hub together with Silkroad Innovation Hub, IT Park Uzbekistan, and Alchemist Accelerator. The program was designed to help B2B startups enter the US market: participants worked with mentors on sales strategy, met investors, attended industry events, and prepared to present their projects at the final Demo Day.
For three months, I spoke with investors and other startups and studied the US market. I saw projects that, in my view, were still pretty rough raise $100,000 or more. It made me realize that what really matters is knowing how to present your idea properly and being able to stand behind what you say.
I saw that I’d been waiting for the perfect moment for no reason. If I could go back, the first thing I’d do would be start selling. I’d build the most basic MVP and say: pay first, and we’ll improve everything else afterwards.
When I returned to Tashkent in December, my partner and I got back to coding.
– How did you decide to change Ucode?
– It turned out that people didn’t need a platform packed with features. They needed a ready-made solution to a specific problem. Over the next two to three weeks, we created templates in Ucode for CRM, HRMS, task management, and financial tools to make the platform easier to navigate. In early 2026, as the number of users started to grow, we decided to add an AI assistant.
“Generating a single project could cost us as much as $1,800”
– How did you integrate AI into Ucode?
– We spent around three months setting up the AI. During that time, we ran into several problems. One of the biggest was token costs. If you use one powerful model for every task, the expenses quickly become too high. I remember one project where we spent almost $1,800 on tokens alone.
So we decided to split the workload between several models. We used Claude for complex planning and architecture, while simpler tasks went to cheaper models, including GPT-5 and Gemini. That helped us bring the costs down.
But that created another problem: the models couldn’t pass memory and context between each other. We needed to make sure that when we switched models, the new AI understood what had already happened in the project and what decisions had been made before. So we started documenting every change. We kept separate records of the project design, the technologies being used, and the guidelines, meaning the rules that couldn’t be broken. Whenever we switch models, the new one reads this documentation first and only then continues the work.
Another issue came up when the AI suggested solutions that didn’t fit a particular market. For example, on one project in Uzbekistan, it recommended integrating Stripe, even though the service doesn’t support payment processing for Uzbek companies. So we started adding separate restrictions and recommendations for different regions.
– What can the AI assistant do for users?
– A user can describe what they need in plain language, and the AI assistant translates it into technical terms and builds the required structure. An MVP can be generated in around 15–20 minutes, although it is not a fully finished product at that point and can still be refined further. The developer can also see the generated code, the database structure, and the relationships between tables, and make changes if needed.
For example, one of our current projects has 163 tables and 293 relationships between them. Even if the AI created all of that, the developer can still see what is happening under the hood and change anything manually. So the AI does not work like a “black box.” The result remains transparent and fully manageable for the developer.
“In the US, I saw that investors don’t have to get in the way. They can actually help”
– How much have you invested in developing Ucode so far?
– We built the project without external funding, using profits from our outsourcing company, Udevs. We never calculated the exact amount, but over time we invested roughly $200,000–300,000. It was money we could have taken out as dividends, but instead we put it back into developing Ucode.
In general, I’m not a big fan of the idea of a project constantly relying on someone else’s money. On top of that, not every investor understands how startups work. Sometimes they start setting their own rules about what you can and can’t spend money on, and what you can and can’t do.
At the same time, I saw a different approach in the US: investors do not have to get in the way. They can actually help. So we are not ruling out investment in the future. For now, whenever we need more money, we simply take on another development project through Udevs, earn from it, and use that income to keep building Ucode.
– How does Ucode make money?
– Before we introduced AI, we charged a flat subscription fee of $300 per month. Once AI was added, we had to change the pricing model because our costs started depending on how many tokens users spent on generation.
We now offer a free plan, which is enough for roughly one generation, as well as paid plans. For example, the basic plan costs $33 per month and includes 200,000 AI credits, which is usually enough to bring a small project to the MVP stage. Users can also purchase additional tokens separately if needed.
We currently have around 20–30 paying companies, including two in the US. Our MRR is about $5,000.
– How did you attract clients in the US?
– We looked for clients through LinkedIn, attended events in Silicon Valley, met potential customers, ran demos, and gave them free access to Ucode.
– How many people are on the team?
– There are currently five people working on Ucode full-time. I’ve mostly stepped away from development myself and now focus on sales and marketing.
– What are your goals for Ucode over the next year?
– In terms of revenue, we really want to reach $10,000 in MRR and break even by the end of this year. Next year, the goal is to raise that to $100,000.
– Which markets do you want to expand into next, and why did you choose them?
– Right now, we operate in two markets: Uzbekistan and the US. We’re not planning to relocate to the US, though. At most, we might open a sales office there for the sales team. We don’t want to move our developers. We’re comfortable in Tashkent, and the cost of living is lower here. The only issue is the time difference, but that can be managed.
We also ran pilot launches in Pakistan, Bangladesh, and India. We attracted around 200 interested users who went into the system and tried generating a project on their own, without any training. That showed us there is demand in those markets, so we want to keep scaling there.