How we built Wynna

Every fact your child practises gets its own review schedule. Get it right and confident? Wynna waits a day before asking again. Right again? Two days, then five, then longer. Get it wrong? It comes back in five minutes. Nearly right? The gap halves. The app automatically surfaces due reviews before new material, so your child always spends time on the facts that are about to fade.

Wynna never shows a fact and asks "did you get that?" Every question requires your child to produce or select an answer from memory: typing it out, completing a sentence from memory, or choosing the right option from four. The engine also distinguishes between "answered from memory" and "answered after a hint" — independent recall earns more XP and counts differently on the parent dashboard, because it's a better signal of real learning.

When your child gets something wrong, Wynna's companion follows a strict three-step hint sequence. First, it asks a question that points them towards the key fact — no answer given. If they're still stuck, it offers a sentence starter or a clue. Only on the third try does it name the specific thing they're getting confused about, explain the principle, and ask them to apply it. The tutor is explicitly forbidden from ever giving a copyable answer. Your child always does the final step themselves.

When Wynna generates revision from your child's material, it identifies the common misconceptions for each topic — the specific wrong answers a child at this level is likely to hold. The AI grader uses these to diagnose what your child actually believed when they answered, not just whether they were right. These misconceptions are tracked with repeat counts. If the same wrong belief keeps appearing, it's flagged on your parent dashboard with a note like "keeps confusing evaporation with condensation (3 times)" — because recurring misconceptions don't fix themselves with more practice.

Every writing task in Wynna is assigned a framework that matches its subject. Humanities uses PEE (Point, Evidence, Explain) — the framework used in UK exams. Science uses CER (Claim, Evidence, Reasoning). Maths uses SOLVE (State variables, Operational formula, Logical steps). Creative writing uses DESCRIBE (Sensory detail, Figurative device, Expansion). The AI coach reads your child's writing, identifies which part of the framework is weakest or missing, and gives them a sentence starter for that section. It never writes the answer for them — it tells them where to focus.

Every typed answer goes through a three-verdict grading system: correct (the meaning is fully expressed, even in simpler words), partial (the right idea but incomplete — a key detail is missing), or incorrect. For partial answers, Wynna tells your child exactly what they got right and what they left out. They get some credit, and the question comes back sooner than a correct answer but later than a wrong one — so they spend time on the specific gap, not re-doing the whole thing. On your dashboard, you see what they understand and what they keep leaving out.

Each topic in Wynna moves through clear stages: Not started → Needs help → Developing → Secure. A topic isn't marked as "Secure" until your child has answered consistently well across at least four graded attempts with 80%+ accuracy. When they open a new session, weak topics and unseen material are always offered first — the engine doesn't let them coast on topics they've already nailed.

Wynna offers several mission modes. "Topic training" focuses on one area (useful when first learning something). But "Mixed mission" deliberately shuffles questions from every subject and topic together. "Weakest areas" interleaves the three topics your child struggles with most. This means every session includes the productive confusion of switching between question types — the same challenge they'll face on exam day.

After getting a question right, your child rates how confident they were. This directly controls the review schedule: a confident correct answer gets pushed further into the future, but a "lucky guess" comes back soon. This prevents children from cruising through easy material while being quietly shaky on the hard bits. On your dashboard, you see whether answers are coming from memory or from hints — two very different signals about how secure the learning really is.

Your child picks a world — Heroes or Mermaids — and unlocks companions by answering questions. Each companion has their own personality and voice, levels up alongside your child, and gives better hints as they grow. The only way to unlock the next one is to revise. This is deliberate: the reward is powered by the learning itself. There are no timers, no punishing streaks, no notifications nagging them to come back, no nudge techniques to keep them online — we're designed around the UK Children's Code, and we take that seriously.

You upload your child's actual exam notes, textbook pages, or photos of worksheets. Wynna reads the material and builds the revision from it — their topics, their syllabus, their exam. Every question, every model answer, every misconception is drawn from what you supplied. The system is explicitly told: "Use ONLY the supplied material. Do not invent topics or facts beyond it." So if the notes are thin, the campaign is thin — it never pads with made-up content.

Your parent dashboard doesn't just show numbers. It computes which topics are weakest and why — "understands photosynthesis but keeps mixing up respiration" — and gives you 2-3 specific things to do tonight. If a wrong belief keeps appearing, you're told about it with a repeat count. If your child is relying on hints instead of answering from memory, you're told that too. The AI-powered Parent Coach turns all these signals into plain English: "Quiz her on the water cycle at dinner — ask 'What is condensation?' and listen for 'when water vapour cools and turns back into liquid.'"

Every piece of content passes through four layers of checking. What your child types is screened before it reaches the AI. What the AI replies is screened before your child sees it — anything flagged is replaced with a calm, safe message. The notes you upload are screened before they become a lesson. And the lesson itself is screened one more time before it's saved. If the safety service is ever unavailable, the whole request fails — we never show unmoderated content. Your child has no account of their own, no login, no public profile, and no contact with other users. Their data is not used to train AI models.