Inspiration: Manufacturing workers are retiring, taking decades of expertise that was never written down. New hires often spend 6+ months shadowing veterans just to learn what “sounds right” or “feels off” on a machine. The most valuable knowledge isn’t in manuals. It’s tacit. Sensory. Experience-based.
We asked a simple question: What if you could capture all of that from a single training video and turn it into something a new hire could learn from on day one?
That’s why I built DayOne.
What It Does: Upload one training video, and DayOne instantly generates a complete Duolingo-style micro-learning course, no content authoring, no instructional design, no manual work.
An AI mentor that answers questions with timestamped citations back to the original video Every learning card includes the real frame from the exact moment in the training video where the concept is taught. It turns passive video into active, structured, measurable learning.
Challenges We Ran Into
- Structured AI Output at Scale Getting Gemini to output consistent, schema-valid JSON across 15 different challenge types was the hardest part. Each card type has unique required fields, and all must be contextually derived from one video.
We solved this with strict Zod schema validation and careful prompt engineering that enforces mandatory content minimums.
- Client-Side Frame Extraction Seeking to exact timestamps and reliably capturing decoded frames into canvas was surprisingly tricky. Different browsers handle video decoding timing differently.
We built a custom seek-and-wait pipeline that guarantees accurate frame capture.
- Simulation Accuracy vs. Engagement Interactive elements like dial calibration and wire-connect exercises need to be both engaging and technically accurate. Generating realistic tolerance ranges and wiring sequences required iterative prompt refinement.
Accomplishments That We're Proud Of
Turning a single training video into a complete structured micro-learning course Extracting tacit knowledge — not just transcription Building 15 interactive challenge types generated automatically Achieving schema-safe AI output suitable for production UX Implementing reliable client-side video frame capture Delivering offline-first functionality
Most importantly, we moved beyond an “AI demo” into something that behaves like a real product.
What’s Next for DayOne
Deploy live for manufacturing teams to test with real training videos Support multi-video onboarding programs Add spaced repetition scheduling to optimize long-term retention Build team analytics dashboards to identify workforce knowledge gaps Expand beyond manufacturing into skilled trades and industrial training DayOne turns retiring expertise into scalable, interactive knowledge — starting from a single video.
Built With
- elevenlabs
- framer-motion-for-swipe-gestures-and-drag-to-reorder-animations
- google-gemini-2.0-flash-for-multimodal-video-analysis
- indexeddb-for-client-side-video-storage
- next.js-14-(app-router)
- supabase-for-cloud-sync-and-authentication
- tailwind-css-for-styling
- three.js-and-react-three-fiber-for-interactive-3d-equipment-models
- typescript
- zod-for-strict-ai-output-schema-validation
Log in or sign up for Devpost to join the conversation.