Acu-Primer

💡 Inspiration

I've always struggled to engage with long-form text material. It's easy to zone out, lose focus and fail to retain key concepts. However, active domains like math and programming keep me fully engaged because they demand direct participation.

I want to bring that same hands-on dynamic to traditional books. This sort of active, interactive learning is strongly supported by research as an effective method for retention. With recent advances in AI, transforming dense books into chunked, interactive learning experiences is now practical.

🛠️ What it does

acu-primer is an MVP that converts static books into interactive, structured courses using an author-in-the-loop approach: AI handles the heavy lifting of content conversion, while authors or educators retain editorial oversight.

• ✏️ Author-in-the-Loop Editor: Instead of fully autonomous generation, the platform centres the educator. Teachers can review AI-drafted modules, provide feedback or specific instructions, and have the AI iterate on the lesson content before it ever reaches a student.
• 🧩 Bite-sized Modules & Active Checks: Lessons are broken into digestible modules. A module can include an interactive exploration (predicting, then stepping through or running a concept) and will feature at most one auto-graded understanding check at the end.
• 🎯Targeted Remediation: The AI pre-drafts fallback help text based on anticipated misconceptions for planned wrong answers. The author reviews and signs off on this help text ahead of time, ensuring students receive accurate guidance when they make a mistake.
• 📊 Instructor Telemetry: Lecturers get access to granular progression logs: the system records each answer given, the misconception it matched, the help shown, the student's predictions, and the time taken between steps. This lays the foundation for deep personalisation.

🏗️ How we built it

• Development Workflow: Used Claude Code with the Devpost skills pack. Claude asked and we shaped the planning documents together. Claude wrote the code, while I drove the scope and reported any issues.
• Frontend: Built with TypeScript and Svelte for a lightweight, responsive client interface.
• Backend & Edge Infrastructure: Deployed on Cloudflare Workers backed by Cloudflare D1 for fast relational data access.
• AI Engine: Powered by the Claude API (Opus 5.5) to draft lesson modules and iterate on instructor edit suggestions.

🚧 Challenges we ran into

• API Budget Constraints: Operating within a tight $20 API budget limited full-scale batch generation across entire multi-chapter books (~$1 per lesson).
• Agentic Edge Cases: Working closely with Claude Code introduced subtle implementation bugs and system edge cases, requiring rigorous manual testing across all core user journeys.

🏆 Accomplishments that we're proud of

• End-to-End Loop: Successfully shipped a live, functional authoring pipeline that ingests source material, drafts modules, and updates seamlessly based on human feedback.
• Resourceful Delivery: Shipped the MVP in just 4 weeks, spending roughly 8 hours per week. This is mostly thanks to the spec-driven workflow adopted.
• **Structural Flexibility*: Successfully parsed and generated interactive courses from books with different structural formats.

🧠 What we learned

• Upfront Context and Shared understanding with the coding agent matters: Investing time in formal specification documents (PRDs, tech specs) directly with coding agents dramatically elevates the baseline code quality and architectural decisions.
• AI as an amplifier: AI-generated lessons are only as strong as their foundation. High-quality source material yields high-quality interactive modules, making this platform a multiplier for expert authors and educators rather than a total replacement.

🚀 What's next for acu-primer

• Cost Optimization: Refine prompt architecture and caching strategies to lower generation cost per lesson.
• Expanded Interactivity: Introduce richer interactive canvas/sandbox widgets for different types of subjects.
• Authentication & Multi-Tenancy: Implement secure user accounts, role-based access, and self-serve course creation tools for authors and educators.
• Actionable Analytics: Leverage student execution logs to supply teachers with actionable cohort insights and individual intervention triggers.
• Offline Accessibility for Underserved Learners: Generate lightweight, offline-first educational content optimised for low-end mobile phones. Because generation occurs once on the platform, entire curricula can be transformed into high-quality, zero-cost offline courses for underserved students.

Built With

• claude
• cloudflare
• ffmpeg
• playwright
• svelte
• typescript
• wrangler