How this Entrepreneur Built the Architecture for Rapid AI Experiments Founder Pritesh Kadiwala breaks down how these engineering decisions became the foundation for Wishroll's speed, scale, and ability to shape new digital experiences for Gen Z.

Wishroll designs products for Generation Z at the intersection of fandom culture and artificial intelligence. In recent years, the team has launched several apps — most notably Kiwi and Status — which together have drawn more than 4 million users. But what sets the company apart isn't just the number of products. It's the way they're built: a modular architecture, ultra-fast iteration cycles, and an approach to AI that slashes costs while enabling bold new experiments. Founder Pritesh Kadiwala breaks down how these engineering decisions became the foundation for Wishroll's speed, scale, and ability to shape new digital experiences for Gen Z.

From Kiwi to Status: From Minimalism to a Complex Ecosystem

Wishroll's story began with Kiwi, a lightweight music-sharing app. Technically, it was simple: a minimal backend, push notifications for new tracks, and a feed where users could exchange songs. Yet its cultural impact was significant. Within three years, Kiwi had been downloaded more than 2 million times and reached the top spot on the iOS App Store in Spain (January 2023) and France (August 2022).

Status, by contrast, was a leap into much more ambitious territory. It's a full-scale AI-driven social ecosystem, reminiscent of Twitter, except every user except you is an AI character. People can interact as themselves or slip into fictional personas, posting, liking, disliking, even facing "cancellation" or climbing the ladder of popularity. The familiar Twitter-like interface lowered the learning curve, and engagement was striking: on average, users spent about 90 minutes a day in the app.

Building something on that scale required a completely different engineering foundation and new strategies for scaling.

A Modular Stack and the Speed of Launch

At the core of Wishroll's success is a lightweight modular stack that lets the team reuse key components across products. The result? Launch cycles measured in weeks, not months.

Each module — from user feedback tools to backend services to AI response generators — is designed to function independently. That means a failure in one piece doesn't bring the entire product down. In 2024 alone, the team shipped eight different applications built on top of Status, combining and recombining existing modules to accelerate development.

Crucially, Wishroll's feedback tools are baked into the architecture and connected across every app. User reactions can be tracked in real time, giving the team clear signals on whether to scale a product or shut it down quickly before costs pile up.

How Engineering Decisions Enabled Scaling

Status's architecture is built around microservices: separate prompts for AI response generation, isolated management of character states, and independent pipelines for processing interactions. This separation allowed the platform to scale to half a million active users at once without major failures.

The modular design also enabled continuous experimentation without destabilizing the system. New mechanics, updated AI behaviors, or additional services could be slotted in without forcing downtime or massive rewrites.

Equally important was Wishroll's disciplined approach to product shutdowns. As Kadiwala puts it: ""If you don't stop projects at the right time, teams fall into the trap of adding new features hoping to fix the product, when the real issue is that the core experience doesn't work. Our framework is to clearly identify the moment a product should be closed and move on. Combined with modular architecture, this mindset lets us make fast decisions and rebuild when needed."

Cutting AI Costs Without Sacrificing Quality

Another engineering breakthrough was reducing AI operating costs by nearly 20x — a critical factor for consumer-scale apps. Wishroll achieved this by embedding a feedback loop where users rated AI responses directly. Those ratings became a ground-truth dataset, which the team used to optimize prompts.

That dataset was then shared with partners at InWorld AI, who helped refine prompt templates and select smaller, more efficient models capable of delivering the same quality at a fraction of the cost. Wishroll also implemented dynamic switching: if a response rated poorly, the system could escalate to a stronger model. In some cases, expensive models were used to generate benchmark outputs, which cheaper models were later fine-tuned to replicate.

This layered approach kept quality consistent while slashing operating expenses.

How Technology Shapes User Experience

While Kiwi and Status addressed different needs, both tapped into the habits and expectations of Gen Z. Kiwi gave users a casual, authentic way to keep in touch through real-time music sharing. Status leaned on the structure of a familiar social network structure, embedded in fandoms and enhanced with AI characters.

Underlying both products is a cultural insight: Gen Z is rejecting the "staged" perfection of older networks built around likes and polished profiles. They're gravitating toward platforms that value authenticity and participation in the moment.

BeReal is a well-known example. Between 2022 and 2023, its Gen Z user base more than doubled. The app's once-a-day photo prompt forces simultaneous shots from front and back cameras, while removing glossy filters and follower counts altogether. The appeal lies in unfiltered moments — a sharp break from airbrushed feeds.

Kadiwala sums up the shift: "Unlike past generations that focused on one or two big networks, Gen Z spreads attention across many specialized apps — TikTok for entertainment, Twitch for streaming, Discord or Geneva for group chats. Increasingly, they don't just want to watch content, they want to create, role-play, and engage in tight-knit micro-communities."

Next Challenges: Narrow Models and New Formats

Large-scale language models are still too costly to sustain for mass consumer products. The next phase, Kadiwala argues, will be highly fine-tuned models designed for very narrow, specific tasks. The smaller the scope, the greater the efficiency and cost savings.

This trend is already visible in the rise of vertical AI models and single-function AI agents. Wishroll is aligning with this shift, seeing it as the most viable way to keep expenses low while ensuring stability at scale.

The company is also exploring new user formats. One area of interest: user-generated 3D fandom worlds, similar to Minecraft or Roblox, where players can build spaces, attend live events, and interact as avatars.

"Technologically, we're close to making this kind of shared fandom space possible," says Kadiwala. "But profitability and user readiness are still big questions. It's an exciting experiment, though not something we're betting on at scale just yet."

Architecture as a Competitive Edge

Wishroll's story illustrates that in consumer AI, success depends not just on creative product ideas but on the underlying architecture. A lightweight stack, modular microservices, built-in feedback loops, and flexible model management allowed the company to launch eight apps in a single year, support hundreds of thousands of concurrent users, and cut AI costs dramatically.

For entrepreneurs, the lesson is clear: in today's consumer market, architecture isn't just an engineering detail. It's a competitive weapon. The speed of experimentation, the ability to adapt to cultural shifts, and the resilience of the business all start with how the technology is built.