🤝 Top Collaborators
🤓 Latest Submissions
Type2Learn — Adaptive Learning for Every Mind
Type2Learn is an accessibility-first nonprofit learning platform for neurodivergent learners, built as a functioning product during the hackathon. Before August 3, Repo/web existed only as an outreach/research landing page used to recruit interviewee; the interactive application was built during the hackathon, with ~80% created using native.builder. The repository and demo evidence the core system directly: deterministic adaptation, objective-level assessment, Speechmatics integration, privacy controls, model routing, and authored fallbacks. There is no single “accessible mode.” Learners choose English or Urdu, layout, colour, spacing, motion, sound, read-aloud, mascot support and how they participate: read, recall, type, speak or apply. With explicit consent, Type2Learn’s adaptive system uses minimal aggregate signals such as reading time, pauses, rereading, task initiation and assessment evidence. Deterministic rules decide whether support is warranted, and the system can propose one reversible change at a time. It does not diagnose learners, silently build profiles or automatically change preferences. AI is Gemini-first, with server-side OpenAI specialist roles (GPT-5.4 Nano, GPT-5.4 Mini and GPT-5.1). Outputs are schema-validated, rule-checked and backed by authored fallbacks. Speechmatics provides authenticated speech-to-text and text-to-speech with usage controls and remote transcription deletion. Adaptive assessments are objective-linked rather than score-led: typed, spoken or MCQ responses identify weak or uncertain objectives and route learners back to the exact content needing review before reassessment. Learners see next steps, not rankings or percentages. Type2Learn was shaped through formative co-design with neurodivergent students, lecturers/researchers and a psychology professional. Early outreach generated 8.3K active users and 23K page views with supporting evidence in the Interviews repository. Built with learners, not merely for them.
10 Aug 2026
Debate Club
Debate-Club offers a complete debate flow: players join queues or private lobbies, topics are suggested by Gemini with veto and refresh options, and roles are assigned by coin toss. Timers, moderation, and turn control ensure fair play. Its AI judging pipeline scores each argument through structured stages and penalizes missed turns. All arguments are embedded and saved in Qdrant for cross-session memory. Users get clear results, feedback, and a PDF summary. The system mirrors agentic-AI design: topic generation, judging, and memory operate as separate agents—similar to orchestration in IBM watsonx Orchestrate. It showcases real-time collaboration between generative models, vector memory, and workflow engines.
19 Nov 2025
