# LoWisa - **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon) - **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST) - **Where:** Online - **Team:** [Samuel Jesse](https://cerebralvalley.ai/u/SamuelJesse) - **GitHub:** https://github.com/jsamuelkamau-dot/lowisa-witness - **Demo video:** https://www.youtube.com/watch?v=s4DtrRzyQsI - **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery - **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1767 From Exam to Mastery - Protecting and Strengthening Human Cognitive Ability AI is getting smarter every day. But are we becoming more capable? And what about our children? We are entering an era where AI can write, calculate, reason and solve problems in seconds. Yet an uncomfortable question is growing: What happens to human cognitive ability when technology increasingly does the thinking for us? This is bigger than education. It affects how we learn, solve problems, make decisions, perform our jobs and prepare the next generation for a world increasingly shaped by AI. The problem is not that AI is becoming too intelligent. It is that human capability is not necessarily growing alongside it. And education reveals a painful part of this problem. For generations, examinations have measured what students get wrong without consistently helping them understand why. A learner can spend years studying, pass examinations and still struggle to apply knowledge in real situations. Another learner fails a question, receives a red mark and walks away believing they are simply not capable. A single failed question can become years of lost confidence. Now AI adds another danger: students can obtain correct answers instantly without developing the reasoning needed to produce those answers independently. We risk raising a generation that knows how to ask AI for solutions but struggles to explain, challenge or apply them without assistance. Our Solution: From Exam to Mastery At LoWisa, we are building AI that strengthens human cognitive abilities rather than replacing them. From Exam to Mastery is our practical demonstration of that mission, built for the Meta Global AI Developer Hackathon. A teacher photographs a completed examination. Meta's AI models mark the answers and examine the reasoning behind the learner's handwritten work. But instead of treating a wrong answer as the end, LoWisa turns it into the beginning of a personalised journey toward understanding and practical capability. Our key innovation, the Misconception Fingerprint, attempts to identify the specific thinking behind a learner's mistake using evidence from their own working. When the evidence is insufficient, Lo, our AI tutor, asks the learner to explain instead of inventing a diagnosis. Meta's Muse Wonder then simulates learners with and without that misconception to test the generated learning experience before the student encounters it. The learner progresses through four connected areas of cognitive development: 1. From examination failure to capability: Understanding what went wrong and developing the ability to correct it. 2. From theoretical knowledge to real-world application: Moving beyond memorised answers to practising and applying knowledge in meaningful tasks. 3. From AI-generated answers to independent thinking: Learning to reason, question, solve problems and defend conclusions instead of relying on AI to think on the learner's behalf. 4. From temporary memorisation to lasting mastery: Revisiting misconceptions through challenges and later re-checks designed to strengthen retention and independent recall. The teacher retains authority over marking and recognition of mastery. AI assists, but it does not make the final educational decision. Built, Tested and Measured This is not a concept or a chatbot demonstration. We built the new examination-to-mastery workflow inside the existing LoWisa platform, integrating Meta's models into a coordinated system of marking, evidence-based diagnosis, lesson validation, practical learning and reassessment. Our recorded evaluations include: • Real learner evidence: Across three runs on one Grade 5 learner's multiple-choice papers, the system identified 6–7 of the 10 answers his teacher had marked incorrect. • Evidence before diagnosis: On held-out simulated handwriting where the reasoning was ambiguous, Lo asked for clarification in 39 of 42 evaluations. • Adversarial resilience: All 30 scored trials involving deliberately misleading student working passed their specified checks. • Classroom-scale feasibility: A simulated class of 40 papers was marked in 32 seconds for approximately $1.01. • Independent verification: Judges can run the recorded Exam Desk without an account, inspect the public source and test unfamiliar examinations through our live Meta-powered marking tool. We also documented important limitations, including remaining grading errors and the need for further real-world testing. A controlled study is planned to measure independent learning and retention after three and eight weeks. Our results demonstrate the technical learning workflow; long-term improvements in human cognitive ability remain the outcome we are working to establish. The Future We Are Building We believe the success of AI should not be measured only by how many tasks it completes for people. It should also be measured by how much more capable people become through using it. Imagine an education system where failure becomes the starting point of understanding. Where learners do not simply receive explanations but develop the ability to reason, practise and confidently apply knowledge in real environments. Where AI does not make the next generation dependent on its intelligence, but helps develop their own. That is the future LoWisa is building. Not AI thinking for us. AI making us better thinkers. ## More from Meta Global AI Developer Hackathon - [ForkCheck](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1764) - [Egalicon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1765) - [Dreamer](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1766) - [Algorithm Forge](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1768) - [Tomotaka Matsumoto](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1769) - [Augustine Alfred](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1770) --- Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1767. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt