ELI5 AI

What I built

ELI5 AI is a Kotlin Multiplatform study companion for university students. It connects notes, active recall, AI explanations, and planning instead of treating them as separate apps

A student can capture a note and turn the same material into an interactive mind map, quiz, flashcards, translation, Feynman practice, or an AI Coach conversation. Plan combines schedules and tasks so the next study action stays visible

Why I built it

Students already have plenty of tools for collecting information. The harder problem is converting that information into understanding and consistent practice. I wanted one workflow that moves from capture → explanation → recall → planning

How it works

The Android and iOS apps share Kotlin Multiplatform domain, data, and Compose UI code. SQLDelight provides local storage and Appwrite handles cloud services. AI requests go through an Appwrite Cloud Function, keeping provider credentials off the device

What changed during Build Week

The project started from the Apache-2.0 StudyAssistant foundation. During Build Week I used Codex with GPT-5.6 to meaningfully extend it with:

1. Production Appwrite setup and a deployed secure AI proxy
2. Connected AI learning workflows and note-to-study-tool entry points
3. The ELI5 AI product identity, Axie Scholar mascot, and app icons
4. A shared Plum Scholar / Liquid Glass design system
5. Redesigned onboarding and four-tab navigation
6. Unified Learn, AI Coach, Plan, and Profile experiences
7. Cross-platform build hardening and iOS/Android verification

The repository and dated history clearly separate the inherited foundation from the work added during the event

How I used Codex and GPT-5.6

Codex worked as an engineering partner across the stack. It audited unfinished migration work, traced KMP dependency and capability issues, implemented UI and platform changes, configured Appwrite, deployed and tested the server function, and repeatedly built both targets to catch real packaging and runtime problems

GPT-5.6 was most valuable when reasoning across the large module graph and preserving existing domain behaviour while changing presentation, infrastructure, and platform integrations. I made the product decisions around the student audience, Feynman workflow, Education positioning, navigation, visual direction, and Build Week priorities

Challenges

The hardest part was coordinating one shared product across iOS, Android, Appwrite, a WebView-based mind-map renderer, and an inherited architecture without breaking existing data flows. Build verification exposed issues that source-level review alone did not, including platform capability contracts and packaging configuration

What's next

The next focus is measuring which active-learning flow creates the strongest repeat usage, improving mastery feedback, and making the AI Coach cite and reuse a student's own learning history

Built With

• appwrite
• codex
• compose-multiplatform
• decompose
• gpt-5.6
• kotlin-multiplatform
• ktor
• sqldelight