01Context
Learning how an EdTech product feels
Wonjun's path to Maieutic AI started with an internship at Letitu, an education technology company. His assignment was to study the user experience of The Pond, a service where students answer a college survey and receive reach, target and safety school matches. He built a review around six criteria: color and font, layout, text balance, readability of questions and choices, quality of recommendations, and error handling. The exercise taught him that in education products, trust depends not only on what the system recommends but on whether users understand why.
02Context
What the review found
The Pond scored well on readability: blue-on-light-blue questions, highlighted target schools, a progress bar and ten focused survey questions made it easy to use. But Wonjun flagged real weaknesses. Some pages failed to load during sign-up, and certain filters took more than ten seconds. He backed this up with research showing that bounce rates rise sharply with load time. He also noted users could not tell why a school was labeled reach or target, or what a 'My chances' percentage meant, and suggested showing which preferences to improve to reach a dream school.
- Recommended cookie-based suggestions on the first screen
- Asked for clearer explanations of chance percentages
- Called for server optimization and better loading indicators
03Problem
Why schools block AI
Later, with three teammates under the name Entropy, Wonjun asked a sharper question: why are schools and parents so intent on keeping students away from AI? Teachers demand 'GPT-killer' reports, and parents see AI as more harmful than helpful. The team's answer was that no AI had been built solely for education. General chatbots simply hand out answers, and because they are probabilistic, they keep talking even when a conversation turns toward content that could threaten a student's mental health. Education-focused tools exist, but most are tied to one company's curriculum.
04Design
Answering curiosity with questions
Maieutic AI takes its name from the Socratic method, where a teacher draws out understanding through questions. When a student asks something, the tutor replies with guiding questions, and as the student answers step by step, they reach the solution themselves. The AI acts as a pacemaker rather than an answer key. It is meant to cover essentially every school subject, to offer support with emotional concerns as well as academics, and to make learning more enjoyable instead of slower. The team positioned it as a student-focused tutor that reduces passive learning and helps guardians notice warning signs.
05Build
An MVP built for students, parents and teachers
The team built the product with React on the front end and FastAPI on the back end, using the GPT API to validate the logic before a planned in-house small model. Guests get an anonymous token and up to ten messages per conversation, lowering the barrier to try it. Registered users keep their full history through JWT authentication. A student can generate a one-time link code to connect a parent or teacher, and role-based access control decides what each sees. Every AI reply is parsed into structured data such as mode, risk level and hint level, and heavy analysis runs in background tasks.
06Design
Supervision without surveillance
Privacy was a top design principle. Parents receive notes on emotional and behavioral signals, such as emotional risk level and dependence on hints, to understand a child's attitude toward learning. Teachers receive focus learning areas and recommended activities drawn from the whole conversation. When the AI generates these summaries, it removes or abstracts names and specific locations, and sensitive counseling content reaches guardians only in summary, so students can talk freely while adults can still spot dangerous negative feelings early.
07Results
What the Conrad judges said
Maieutic AI was entered in the Conrad Challenge's Cyber-Technology & Security category, and the team earned the Conrad Innovator honor. Judges praised the shift from an answer engine to a thinking facilitator, the combination of Socratic dialogue with privacy-aware guardian dashboards, and an MVP backed by real architecture rather than just an idea. They were equally direct about gaps: no external beta data yet, emotional-risk detection not quantified, regex-based parsing that may not scale, no market sizing, and no pricing or unit economics.
- "A strong, well-conceived innovation that excels in purpose, framing, and societal relevance."
- "Most tools just give answers; yours actually teaches."
- "IP defensibility is the weakest part of this category."
08Next
Reaching students who need it most
The team's pilot plan is to give socially marginalized, low-income students free access in exchange for anonymous learning data and feedback, then grow through school communities, parent networks and education NGOs. Revenue would come from tiered subscriptions, a premium plan with multi-care support, detailed reports and nighttime usage controls, and an institutional plan. Looking further ahead, the team wants to build its own small language model and an on-device version for places without reliable electricity, and to add a feature that helps shy students build real friendships instead of depending on AI.
