Suppose you have a 60-page PDF textbook and need 25 multiple-choice questions by Friday. They also need to enter the school's Moodle question bank and be automatically graded. Writing questions is not inherently difficult, but it takes time, and every chapter brings the same work again.
Questgen handles draft generation: put in text, a web address, a PDF, or a DOCX file; choose question types; get questions; edit them; and export formats such as Moodle XML or QTI. Its website lists file formats without claiming to connect directly to Moodle or automatically write into its question bank. That next step is not covered by the public materials.
What makes me pause is the export list[1]: Moodle XML, QTI 2.1 Zip, GIFT, AIKEN, RESPONDUS, and several CSV formats. PDFs have three variants: questions and answers, questions only, or answers only. Moodle XML is generally for Moodle question banks, while QTI 2.1 serves systems supporting a common quiz interchange format. The others leave openings for different banks or tools. These names may mean little to someone unfamiliar with learning management systems. For someone who must deliver, however, the export choice determines whether the grader recognizes the draft. Generation makes drafting faster; export formats determine what kinds of question banks the draft can enter. Whether a system accepts and grades it still depends on the receiving end.
Image: from the product website[1], obtained on 2026-11-18. Original official illustration, not an independently tested result.
The website says it supports single-answer and multiple-correct-answer multiple-choice questions, true/false, fill-in-the-blank, FAQ/short answers, higher-order questions, matching, and quizzes organized by Bloom's taxonomy. Generated questions can be edited and exported, shared publicly, or embedded as interactive quizzes using an iframe. For input, its How it works page[1] says up to 80,000 words of text in any language per run. Another high-volume claim on the homepage says 100,000 words can produce up to 150 quizzes in one go.
One detail places it clearly in the workflow: for turning audio or video into a quiz, the website's example requires downloading a YouTube video as an MP3 before uploading it. It generates questions, but the media file has to be converted first; it does not fetch that video for you. This small handoff shows a tool for moving materials in and out, rather than an “AI teacher.”
An open-source library before a SaaS
In 2020, Ramsri spoke with several edtech companies and found that question writing was often outsourced (original interview[2], 2024-02-11). It was not that nobody could write questions: the work consumed people and never ended. That demand later became the product. The materials do not explain why it offers such a varied export list. The list itself nevertheless shows that different systems have different entry points, and question writers often have to use the system chosen by someone else.
His sequence was somewhat reversed: train his own models, build an open-source library and courses, then move into SaaS. The Questgen.ai repository[3] on GitHub can be installed with pip, includes a Colab demo, and lists multiple-choice, true/false, FAQ, paraphrasing, and question answering in its README. Its last visible commit dates to November 2023. It now looks more like a signboard proving he actually built something than a manual for the hosted version. Public materials do not clearly specify the hosted product's models or data-retention practices. The 2024 interview[2] only identifies LLM APIs such as OpenAI and Anthropic as the largest cost.
Unfamiliar with full-stack development, he built the first version on Bubble.io, later moving to NextJS 13 and React. There is nothing romantic about that choice: getting it running mattered more than getting the architecture right.
Acquisition was slow too. He first wrote 10 to 15 Medium posts, spent about a year building traffic, released a Udemy course, then placed the SaaS link in his posts, course, and GitHub repository. He also built SEO landing pages for specific features. He names blogs, Google search, courses, and GitHub as the main growth drivers, but the materials provide attribution rather than conversion rates or the number of paying users from each channel.
Pricing came before a clear pricing strategy
The original payment model was buying X runs at a time, not a subscription. His explanation was plain: he was not good at implementing subscriptions and added them about a year later. It sounds more like an admission than a strategy retrospective. Many small products grow their pricing this way: first find a form that can take payment, then adjust it over time.
The current pricing page[4], read on 2026-11-18, lists 20 free runs on Basic. Pro is US$15 a month or US$100 billed annually, advertised as unlimited runs under fair use, with priority support and early access to new features. An older official introduction[5] and the 2024 interview both give an annual price of US$150; the older post also mentions a 25,000-word limit per prompt. Today's pricing page says US$100 a year, so the older figures cannot serve as current terms. For the B2B API version, the 2024 interview[2] gives a starting price of US$299 a month. That was the quote then, not a current price list.
Revenue needs care. The interview dated February 11, 2024[2] opens with 60k+ users and US$2,100 MRR, then later gives US$4,000 MRR and 230 paying customers across B2C and B2B. The same article contains two figures without making clear which was current; both remain founder-reported. It also says the founder operated alone with an 85% profit margin, then gives operating costs of about 15–20%, chiefly LLM APIs. Neither passage defines which costs are included or gives absolute costs, so how profit is calculated remains unclear. I am not going to calculate it on his behalf.
Another Indie Hackers interview[6] says Questgen had generated over US$150,000 cumulatively in roughly four years and had over 150,000 registered users. The same article's US$100,000 peak total ARR in 2025, current combined US$6,000–7,000 MRR, and 2.5 million-plus registrations belong to Questgen, Supermeme, and AiArtist together, not Questgen alone.
Plenty of alternatives, but no basis for declaring a winner
Quizgecko[7] is a ready comparison for the same task. It also generates six types of editable questions from PDFs, PPTs, notes, URLs, and YouTube links, and offers PDF export, printing, and shareable links. Some plans support Canvas and Moodle Aiken. Its November 2026 US-dollar pricing[8] includes a limited free plan, Plus at US$16 a month or US$108 annually, and Ultra at US$29 a month or US$276 annually. There is no publicly comparable test using the same inputs, question types, and exports, so I cannot say which generates better questions. The deciding difference may be less about question quality than which formats your system recognizes, and how fully.
Why does someone build this business? The answer is ordinary: question writing is repetitive work, and repetitive work gets split into pieces and sold. Why does someone buy it? I found no buyer interviews, so I can only infer from the export formats: perhaps the buyer wants not “AI to write questions” but a file that can enter a question bank. The model accelerates the middle. At both ends—getting materials in and questions into use—the parts visible in the public materials still leave people making decisions.
[1] www.questgen.ai
[3] github.com
[4] www.questgen.ai
[5] www.questgen.ai
[7] quizgecko.com
[8] quizgecko.com