ShortForm

Inspiration

We have all been there: you spend hours reading a profound non-fiction book, nodding along to every chapter, only to realize a week later that you can barely remember the core concepts—let alone apply them. Passive reading creates an illusion of competence, but real growth requires action and reinforcement.

We built ShortForm to bridge the massive gap between consuming information and actually absorbing it. We wanted to build a tool that shifts the mobile reading experience from a passive scrolling activity into an active, high-retention study system.

What it does

ShortForm is a cinematic, dark-mode-first native Android application that serves as an interactive intelligence layer for your library.

• Instant Book Discovery: Users can search through millions of titles using a lightning-fast global search bar powered by the Google Books API.
• Multi-Persona AI Summaries: Generates concise executive summaries or alters the tone on demand (e.g., standard professional insights vs. an "Explain Like I'm 5" simplified breakdown).
• 7-Day Action Roadmaps: Breaks down a book’s abstract philosophy into a daily, concrete execution checklist with interactive checkbox nodes to gamify implementation.
• Concept Shield (Flashcards): Generates active-recall flashcards featuring sleek 3D-flip animations in Jetpack Compose to test your understanding of key ideas.
• Performance Quizzes: Dynamically generates multiple-choice comprehension checks with instant feedback and structural explanations to lock in retention.

How we built it

ShortForm was engineered from scratch as a modern, performance-oriented native Android app using a robust mobile architecture:

• UI Framework: 100% Jetpack Compose for declarative UI building, smooth micro-interactions, and Material Design 3 components.
• Architecture: Clean MVVM (Model-View-ViewModel) design patterns ensuring clear separation between the UI presentation layers and background data handling.
• The Intelligence Engine: Driven by the Google AI Kotlin SDK utilizing the gemini-2.5-flash model. We utilized strict system instructions combined with setting the responseMimeType to application/json, forcing the model to output a strictly structured, highly complex data tree.
• Data & Persistence: Retrofit handles network handshakes with the public Google Books API, while a local Room Database acts as our smart caching layer. When a summary is generated, it caches instantly to Room, eliminating redundant API hits, keeping costs down, and allowing lightning-fast offline retrieval.
• Image Delivery: Coil manages synchronous high-resolution loading and image caching for book posters.

Challenges we ran into

• Type-Safe JSON Deserialization: Forcing an AI model to strictly adhere to a multi-nested JSON array (containing titles, roadmaps, flashcard objects, and quiz configurations simultaneously) can occasionally result in broken brackets or mismatched fields. We spent significant time debugging our system instructions and implementing defensive parsing using Gson to gracefully catch and handle any malformed AI payloads without crashing the user's view.
• Complex UI States in Compose: Managing simultaneous async states—fetching book metadata from one public endpoint while rendering dynamic shimmer effects for an ongoing Gemini stream—required meticulous state machine architecture within our ViewModels to prevent layout stutter or race conditions.
• Jetpack Compose 3D Rotations: Implementing the 180-degree physical rotation for the Concept Shield flashcards required custom graphicsLayer modifications and camera-distance tweaking to ensure text on the back didn't render mirrored or clip during execution.

Accomplishments that we're proud of

• Zero-Lag AI Processing: By choosing gemini-2.5-flash and pairing it with a local Room DB cache, we built an incredibly snappy user experience where an entire multi-featured study deck generates in under three seconds and loads instantly on subsequent visits.
• Flawless UI/UX Polish: We managed to design a beautiful, cohesive dark theme with bespoke canvas-drawn vertical timelines, fluid animations, and custom loading states that look like a retail-ready product built over months rather than a quick hackathon weekend sprint.
• Highly Optimized Token Management: Crafting an efficient system prompt that extracts text-heavy insights while keeping payload wrappers incredibly lean, maximizing efficiency without sacrificing depth.

What we learned

We gained deep expertise in writing robust, production-grade system instructions for the Gemini API, specifically discovering how enforcing deterministic JSON outputs allows developers to treat LLMs as reliable structured backend engines. We also sharpened our understanding of advanced Jetpack Compose states, multi-layered Room database schemas, and clean architectural design patterns under tight hackathon timelines.

What's next for ShortForm

• OCR Book Scanner: Allow users to use their device camera to snap a photo of a physical book cover or a specific page snippet, instantly pulling up the corresponding AI study suite.
• Social Accountability & Accountability Streaks: Introduce a peer system where users can share their progress on their 7-day action items, fostering group reading challenges and competitive study leaderboards.
• Conversational Author Personas: Leverage Gemini's chat capabilities to let users directly text-chat with a simulated persona of the book's author, allowing them to debate concepts and ask custom contextual questions about the text.

Built With

• ai
• android-studio
• gemini
• kotin
• material-ui