Demostra
AI study app that checks whether you can explain a topic from memory. Live on iOS and Android.
Demostra is an active-recall study app I founded and built on my own. You close the source and explain a topic aloud or in writing; the app marks it proven, blurry or needs review, names what was missing, and schedules the weak topics back before your exam date. It runs on React Native (Expo) for iOS and Android, with a Node.js/Express API on MongoDB, GPT-6 Luna grading, on-device speech recognition, ElevenLabs voices for mock interviews and RevenueCat subscriptions.
Context
AI has made a ready answer available for almost anything. Self-development content keeps multiplying, yet learning is getting shallower: it is easier than ever to consume someone else's explanation, attention is pulled in more directions, and the feeling of having learned quietly takes the place of learning. Reading a summary feels like knowing the topic, right up until you have to explain it.
Problem
Rereading and highlighting create an illusion of knowing, and the gaps only show up on exam day or in the interview. The learning research is clear about what works: retrieval practice (Roediger & Karpicke), a 2017 meta-analysis by Adesope and colleagues across 217 studies, and Dunlosky's review of study techniques all rank active recall and spaced repetition near the top. Yet the study apps for Turkish exams like KPSS, ALES and DGS are built around question banks. None of them ask you to put the source away and explain the topic in your own words.
Approach
I built it end to end. the mobile app, the backend and the website. You set a dated goal, such as an exam, an interview or a presentation, and AI drafts its topic map. You explain each topic in about 40 seconds; GPT-6 Luna grades the explanation; the grading model has to pass my audit of hard cases, which gpt-4o-mini did not. Speech is recognised on the device and audio is never written to disk. A readiness board shows what is proven and what to do next, and deliberately never invents an "85% ready" score. Weak topics come back the next day, blurry ones in two days, and every interval is capped by the goal date. Communities, live group sessions and voice mock interviews sit on top. For institution accounts I set one rule and enforced it with tests: an institution sees learning results on its own curriculum, never an individual's attention data. Per-user AI quotas and spend tracking keep the costs under control, and the project carries 98 test files plus a KVKK/GDPR package with a DPIA.
Outcome
It went from the first commit to Google Play in about three weeks, and was live on both the App Store and Google Play in about five, with subscriptions added in version 1.1. Demostra is newly launched and I have just started promoting it, so these are its first users.
