Techie.Lk recently had a chat with Prasanth Subendran, Co-Founder & CEO of MindMap AI, a home-grown AI product and mind mapping tool, which can turn inputs like PDFs, audio and video into editable mind maps.
Launched in March 2024, MindMap AI surpassed 100,000 global users in roughly one year.
Prasanth, who is also a co-founder of 3Axis Labs, a Jaffna-based software company, that builds software products for clients in Australia, Singapore, Denmark and the UK, has over ten years of experience in the local startup and tech ecosystem.
His mission, along with his fellow colleagues and co-founders, is to increase high-quality IT jobs in Sri Lanka’s North and East.
What problem was MindMap AI created to solve, and what did you see missing in existing AI tools that led you to build it?
We started building the product in early 2024, when the AI boom began. At that time, existing mind-mapping tools had no AI, so we decided to build an AI-first mind-mapping tool. We started as an AI mind map generator in 2024 and later added a mind map editor and other features.
Before LLMs, you had to go through PDFs, books, videos, or meeting recordings, take notes, and then convert them into a mind map. The point of a mind map is to get a mental picture of a topic and to explain it easily to others. LLMs can do the summarisation, so we let users bring any input (PDF, audio, video, meeting recordings, images) and it’s converted into a mind map instantly. People no longer waste time going through long PDFs or recordings, and that’s the value we add.
LLMs are probability-based, not rigid in how they take input and give output. The first generation isn’t hard, since anyone can create a mind map from a PDF or a prompt. The challenge comes when you want to keep evolving or editing the mind map. The AI has to understand clearly what is being asked, and the output must not lose anything already created, including manual changes users have made. We solved this by passing the data to the AI in a structured data format, so it clearly understands the request, and we have a stable way of communicating with it.
We ground the output in the input provided, whether that’s a PDF, audio, or video. The mind map is created only from that input. Otherwise, the AI pulls information from its own model data, which can sometimes be unhelpful.
What is MindMap AI’s strongest differentiator, and why should someone choose it over ChatGPT or other established AI platforms?
Mind mapping is a structure, and we provide editing features. You can move notes around and mark items as done or important. We’re also not competing with ChatGPT. We provide Model Context Protocol (MCP) integration for ChatGPT, Claude, and other AI platforms, so people can use the intelligence of the models they already have and visualise the result in our mind map editor. ChatGPT calls this a plugin and Claude calls it a connector, and we’re listed on both. Those models are very powerful, so we can harness their intelligence.
Productivity has thousands of tools, but mind mapping is a small subset of it. Many older players who offered manual mind maps never integrated AI, and they’re being left out of the market. Others have started adding AI but struggle because their architecture was built for manual mind mapping. Since we started as AI-first, it’s easy for us to offer both AI and manual mind mapping. We offer manual mind mapping completely free, because we see it as a commodity.
In terms of privacy and data security, we use Google’s Gemini model, so data is processed under the security and privacy terms Gemini provides. Our servers are hosted on Google Cloud. We’re also a US-registered company, so we have data security and privacy commitments to users under US requirements. Nobody has access to the data. We don’t have our own model, and we don’t train on user data.
Who is your core target user today? Is it students, professionals, businesses, researchers, educators, or another segment? How has that target evolved since launch?
Our core users are mostly in education: students, researchers, educators, and lecturers. We also have individual professionals who use mind maps in their work, such as coaches and programme managers. Anyone in the planning stage uses mind mapping because it makes information easy to visualise, and once you can visualise it, implementing the plan is easier. Not understanding a topic or idea is the hardest part. So, our target market is mostly people in the planning stages.
What does building an AI product from Sri Lanka look like in practice? What advantages does the local ecosystem provide, and what are the biggest limitations?
This applies to any tech startup, not just AI. Financing is the most challenging part. Somebody has to partner with you to finance the idea. You can bring in co-founders with complementary skills, like a developer if you’re a marketer, but who is going to be the financier co-founder? In my opinion that problem is not solved, and that’s why not many startups come out of Sri Lanka. The ecosystem is helpful in many ways, with mentoring, programmes, implementation, and other support, but financing is the missing piece.
The second challenge is talent. Building an AI product needs solid people who are problem solvers, and that’s true for the whole industry. We have the talent, but they’re working in corporate jobs because they need job security. So high-calibre people aren’t attracted to startup building. That’s again connected to financing. If financing were easier, more people would be attracted.
Compare this with the US, where Y Combinator gives $500,000. Does anyone here give even a fifth of that, $100,000, through an accelerator? Do we have accelerators that bring real investors to a demo day? We don’t. Talent here can earn $2,000 to $3,000 a month, but in Singapore, the UK, or Canada they can earn $5,000 to $7,000. If the money problem is solved, we can attract high-level talent. Building a complex technical product needs smart, intelligent people with a solid builder mentality, and I’m not saying that depends on age or experience.
Our advantages are that, as founders, we take less than market salaries and manage with a minimal team. We can also compete on price in the AI market. That doesn’t mean being the cheapest, because a cheap price tag makes the product seem cheap. We price the same as our competitors.
What role do you see AI playing in education and knowledge work in Sri Lanka, and where do you think local users are still underestimating its potential?
I think AI will play a huge role, and it may already have started. If you have your curriculum, you just ask the AI. What you need is consistency and the mindset to sit and study, while the AI provides the knowledge. You can ask any question and get step-by-step answers, especially with ChatGPT or Claude and especially on the paid versions. You can treat AI as a personal teacher for each subject, though it can’t give you the patience to sit and study.
On local users, I think the potential is being underestimated. AI adoption is coming mostly from Western countries, especially the US, so it will take time here. We only adopted mobile and online payments recently, around COVID. But the use cases of AI are massive across industries. IT is being hit first, with layoffs, because that’s an international kind of job. Other jobs are more local. For knowledge work, AI means we’ll need far fewer people. Physical work still needs people unless robotics takes over.
MindMap AI is a global product, and we didn’t initially target Sri Lankan users. Google Trends shows continuing worldwide demand for mind mapping, especially from the education sector. We do have Sri Lankan users, but students and educators all over the world use it.
As foundation models become more capable and cheaper, is that an opportunity for MindMap AI’s business model?
To be honest, foundation models are not getting cheaper, especially the more capable and complex ones. But there are also smaller models coming up behind, so at some point we may have cheaper ones. We don’t need our own model. We’re not in a market where we’d have to build one, so we can use the existing models. If open and other models get cheaper, the price point goes down, and we can offer customers a better price. So, it’s an advantage, not a threat.
What is the next major stage for Mind Map AI: new AI capabilities, enterprise adoption, international expansion, or something else?
International expansion isn’t a separate plan, since users from all countries already use the product. But we’ll target the North American market more.
Enterprise adoption will be limited, because mind mapping is mostly for individuals and small teams. In education, though, universities and institutions are an opportunity. We can offer bulk discounted licences, say 1,000 or 2,000, so students can use the tool throughout their semesters.
On AI capabilities, as new models become more powerful, the accuracy of the generated mind maps improves.
We’ve also nearly caught up with players who have been in the industry for 20 to 50 years, like XMind. We did this in the last 24 months. We have a few features and milestones left, and then we’ll have a state-of-the-art product competing with 20-year-old products.