How Ramsri Golla Built an AI Product Portfolio After Aurora

From outsourced exam work to winding down Aurora

In mid-2019, Ramsri Goutham Golla was in Singapore with an education-technology founder, looking for ways to apply artificial intelligence (AI) to education. After considering online-class analytics and personalized learning, the two noticed that education companies outsourced exam-question production. If questions could be generated automatically from passages, repetitive authoring work could be reduced, and from that idea they started Aurora, an AI question-generation business.

Aurora received early investment from the startup backer and investor Entrepreneur First, and also lined up an investor to lead a 1 million USD follow-on round. But conflict between the co-founders made it difficult to continue the business. The team shut the company down and returned the remaining investment funds, and Golla returned to India, taking about half a year of rest while preparing for his marriage.

Growing Questgen on freelance income

The life he chose afterward combined freelance work with product development. He took on AI-related contract work and sold online courses while developing the software he wanted to make. In a 2022 interview, he said he was running consulting, education, and a small software business together, a mix of income streams that can be seen as giving him room to wait for early product revenue.

The question-generation idea continued in Questgen, which he developed on his own. The early interface combined an input box for passages with an output window showing generated questions, and he solved web development by learning the no-code tool Bubble. He added payments to it, and later, as more features became necessary, learned web development skills and rebuilt the product.

His technical writing helped him reach customers. Golla distributed question-generation code as open source and attracted interest from developers and the education field through related articles. By adding a link to the web service in the GitHub repository, even people who found it burdensome to install and operate the code themselves could use the product. The public code and explanatory articles became an entry point into the paid service.

Questgen's monthly recurring revenue (MRR) was about 500 USD in a June 2022 interview, and grew to around 3,000 to 4,000 USD in Golla's August 2024 retrospective. He said at the time that the product ran with little maintenance. A first product built over several years had reached a stage where it earned revenue without demanding much daily work.

A second business built by three people

In January 2022, while growing Questgen, Golla started a new experiment with two people he had met online. Sanjeev NC, who made memes, Nico Botha, who developed web services, and Golla, who handled AI features, became the co-founders of Supermeme. Centered on a WhatsApp group chat, the three exchanged work updates in their own time and built the product.

Early Supermeme was a tool in which users picked a topic and a meme image, and AI wrote the caption. The team published results on social media to gauge reactions, and when real paying users appeared, they confirmed its potential as a paid product. It was later launched on Product Hunt, reaching No. 1 Product of the Day and becoming known to more people.

While improving the product, they found that choosing a suitable meme image was itself difficult. They built a search feature so that even people who did not know a familiar meme's name could find images by entering an emotion or expression. They then added text-based generation that handled image selection and caption writing together, and as non-English-speaking visitors increased, expanded support to more than 110 languages.

In February 2023, Supermeme reached 100,000 signups. The three founders spent no money on paid marketing, but their long-running social media accounts had about 100,000 followers combined, with little overlap. After they introduced the product, users shared memes, and accounts covering AI tools introduced the product again. In a retrospective, the team cited the audience the founders had already built as an important asset in early distribution.

In the first quarter of that year, it recorded 150,000 cumulative signups and 10 million memes generated, and early in the second quarter it reached 100 paying customers. Because AI usage bills also rose with usage, growing paying customers mattered alongside gathering visitors. The fact that customers started registering cards and using the product without separate calls suited a team running other work in parallel. The team named meme-quality improvement and expanded search traffic as later tasks.

The extra work that came with more products

For Golla, publishing his development process was a separate job. In a 2023 interview, he said he allocated two hours a day to content creation and repurposed what he wrote for LinkedIn, X, Medium, and YouTube. Even in fields such as education or translation, where he lacked connections, publishing his work could bring customers and collaborators. Alongside time spent developing products alone, he secured time to let people know those products existed.

The product lineup he operated at the time also included Supertranslate, a subtitle-generation tool. It created English subtitles from video or audio in several languages, and Golla continued experiments applying new AI technology to other tasks. He said he used infrastructure billed by compute usage so that even a small product with 100 USD in monthly recurring revenue could operate profitably.

The burden of running several products also appeared in the same interview. He had to manage a separate code repository for each product, and repeatedly switched from a consulting client's development environment to course production and recording. Golla said these context switches were hard and that, long term, he wanted to focus on one product. At the time, though, he enjoyed making each product and continued operating them because acquisition offers were not satisfactory.

The phrase he used to describe his pace was slow-learning founder. Even in a post announcing a business at about 100,000 USD in annual scale, he said it had taken more than five years of part-time work to reach that level. It was the result of developing products in remaining time while maintaining consulting and courses for living expenses, along with parenting and travel. In this choice, time with family and a state in which he could keep enjoying development were considered alongside business size.

Video-generation experiments and criteria for the next product

**An experiment extended into video generation.** AiArtist, later added to the lineup, is a tool that takes text input and creates videos combining animated text and visual elements. Golla centered the product on short-video needs such as social media posts, ads, and product introductions. Users enter what they want to convey, and the service generates the image composition and motion and exports them as a video file. After questions and memes, it covered repetitive work in video production.

While developing AiArtist, he also tested several implementation approaches himself. In an article published in February 2026, he compared building motion with code against using video-generation models. Addressing the problem that text inside videos was not rendered properly, he described starting with an image model that renders text well to create opening and ending scenes first, then passing them to a video model. It was an approach that split generation into multiple stages to improve finished quality.

Pricing was also arranged differently for the product. As of September 2026, AiArtist's launch price was a one-time 99 USD payment for one year of access, with no automatically renewing subscription. During the access period, the balance was refilled to 100 credits every 30 days, and unused credits did not roll over, so the usage customers received and the compute the service would provide were both bounded.

**Criteria for spending time on the next product.** The three products disclosed in a July 2026 interview were at different stages. Questgen had passed 150,000 USD in cumulative revenue, Supermeme had more than 2,000,000 registered users, and the recently launched AiArtist was at an early stage with more than 20 paying users. While existing products earned revenue, the new product was in an experiment confirming payment demand.

Combined monthly recurring revenue across the products was then around 6,000 to 7,000 USD. Since the 2025 peak annualized recurring revenue (ARR) Golla disclosed was 100,000 USD, adding more products did not lead to continued revenue growth. He cited stalled visitor and subscription growth as large AI services offered related features directly as an important difficulty.

His response was to examine how customers used outputs in real work. Questgen offered editing of generated questions and export in formats used by education systems, such as QTI and Moodle XML. Golla described these input and output formats as differentiators from general-purpose AI services. His judgment was that detailed features that completed customers' work could remain a reason to keep using the existing products.

Golla's experimental method in the 2026 interview was to publish a small prototype first and spend more time on ideas with good responses. Showing a short demo or content from testing a new model could confirm people's interest before full development. In his case, the advantage of making several products was being able to repeat this selection. He kept narrowing where to invest the next few years while watching existing products' revenue and operating burden.