Jarvis

Inspiration

"We are what we repeatedly do. Excellence, then, is not an act, but a habit." — Aristotle

I started with a simple observation: 92% of New Year's resolutions fail. Why? Because most habit tracking apps are passive. They are essentially just digital spreadsheets that require you to do all the heavy lifting—planning, scheduling, and remembering.

I wanted to build something that wasn't just a tool, but a Butler. A true butler doesn't just hand you a to-do list; they organize your day, remind you discretely, and adapt to your needs. That was the spark for Jarvis: an agentic application that moves from "Passive Tracking" to "Active Coaching."

What it does

Jarvis is an AI-powered habit tracking and coaching app that serves as your personal digital butler for self-improvement. Unlike generic habit trackers, Jarvis:

🎯 Learns Your Goals - During onboarding, Jarvis understands your personal improvement goals and creates a customized habit plan.

📅 Schedules Intelligently - Based on your wake time, sleep time, and daily availability, Jarvis auto-generates an optimal daily schedule.

📚 Tracks Reading - Log books, track pages, monitor reading streaks, and (coming soon) take AI-generated comprehension quizzes.

💪 Monitors Fitness - Log workouts, upload progress photos, and get personalized workout suggestions.

🤖 Acts Autonomously - Don't just chat—command. Tell Jarvis to "Add a morning run" or "Change my reading time", and he modifies your database automatically using Agentic Tool Calling.

📊 Visualizes Progress - Beautiful dashboards with progress rings, streak indicators, and achievement badges.

How I built it

Jarvis is a showcase of the power of Flutter + Serverpod working in harmony. By using Dart across the entire stack, I achieved a level of type safety and developer velocity that felt like magic.

The core architecture relies on a specialized Agentic Loop:

$$ Action*{agent} = f(Goal*{user}, Context*{time}, History*{habits}) $$

The Brain (Serverpod + Gemini): I implemented a provider-agnostic AI service that processes user goals and availability.
The Body (Flutter): A beautiful, glassmorphic UI that uses Riverpod to react instantly to state changes.
The Nervous System (Serverpod Client): The auto-generated client ensured that my frontend and backend were always speaking the exact same language.

Frontend (Flutter)

Riverpod for reactive state management
Firebase Auth for user authentication
Firebase Storage for image uploads
Custom design system with dark mode
Feature-first architecture

Backend (Serverpod)

Type-safe API endpoints
PostgreSQL database with auto-generated migrations
Modular AI service abstraction
Docker-based development environment

AI (Google Gemini)

Habit recommendations based on user goals
Daily motivation messages
Function Calling (Tools) for autonomous database actions
Voice-enabled chat coaching

Challenges I ran into

The journey wasn't without its hurdles. My biggest challenge was Conquering the Serverpod 3 Migration. Adapting to the new relation syntax and ensuring my database queries were optimized took significant trial and error. Specifically, mastering the explicit relation handling for our complex Habit $\leftrightarrow$ HabitLog relationships required a deep dive into the new ORM capabilities.

Another challenge was Real-time Synchronization. I wanted the dashboard to feel alive. When a user completes a habit via voice command, the UI needs to update instantly. Solving this required a robust state management strategy with Riverpod to ensure data consistency across the app.

Serverpod 3.x Migration - Adapting to new relation syntax and enum handling in Serverpod 3.
State Management Complexity - Designing Riverpod providers that efficiently handle loading, error, and data states while staying reactive.
AI Integration Architecture - Building a flexible AI service that can swap providers (Gemini, OpenAI, etc.) without changing business logic.
Real-time Updates - Ensuring the dashboard reflects habit completions immediately across all widgets.

Accomplishments that I'm proud of

✅ Complete Flutter + Serverpod Stack - Full working app with backend, frontend, and AI integration.

✅ Flexible AI Architecture - Provider-agnostic AI service that supports multiple AI models.

✅ Beautiful UI - Dark mode design with smooth animations and glassmorphism effects.

✅ Comprehensive Features - Reading tracker, exercise tracker, schedule generator, and AI chat.

✅ Clean Architecture - Feature-first Flutter structure with proper separation of concerns.

What I learned

The most profound realization was the power of End-to-End Type Safety. In a traditional stack, modifying an API endpoint is a scary refactor. With Serverpod, it became my favorite feature. I could rename a database column in the backend, run the generator, and instantly see where the frontend code needed updates.

I also learned that AI Agents are the future of app interaction. By giving Jarvis "Tools" (function calling) instead of just text responses, I turned a chatbot into a powerful operator that can actually do things for the user.

Serverpod's Power - The auto-generated client and type-safe APIs dramatically reduce boilerplate.
Riverpod Patterns - StateNotifier with FutureProvider provides excellent async state management.
AI Prompt Engineering - Crafting prompts that return structured JSON for programmatic use.
Full-Stack Dart - The productivity of using Dart on both frontend and backend.