Aedar | Devpost

Logo
Landing page
Chat page
Generated Roadmap

Inspiration

Every developer knows the pain: you have a brilliant project idea, but turning it into an actionable plan feels like a project in itself. We spend hours breaking down tasks, estimating timelines, and trying to fit everything into our already-packed schedules. By the time we finish planning, we have lost the momentum to actually build.

We asked ourselves: What if AI could handle the entire planning phase? Not just generate a to-do list, but create a structured technical roadmap with phases, milestones, resources, and then automatically schedule it into your calendar.

That question became Aedar -- an AI execution agent that bridges the gap between "I have an idea" and "I am shipping it."

What it does

Aedar transforms vague project ideas into structured, actionable roadmaps through a conversational AI interface. Here is the workflow:

Describe your project - Tell Aedar what you want to build in plain English (e.g., "Build a fitness tracking app with React Native")
AI generates your roadmap - Aedar analyzes your request and creates a comprehensive technical roadmap with:
- Multiple phases (Setup, Core Development, Testing, Launch)
- Detailed milestones with time estimates
- Curated learning resources for each step
- Adjustable planning depth (Sprint, Standard, or Architect mode)
Auto-schedule to your calendar - Connect your Google Calendar, and Aedar finds open slots in your schedule to book each milestone automatically
Track and iterate - Save roadmaps, mark milestones complete, and modify plans as your project evolves

The platform includes three planning modes:

Sprint (Concise): 3-5 key milestones for simple tasks
Standard (Balanced): Full architecture, development, and launch phases
Architect (Deep): Detailed breakdown with risk analysis, tech stack recommendations, and QA steps

How we built it

Frontend (Next.js 15 + TypeScript)

Built with the App Router for server-side rendering and optimal performance
Neo-brutalism design system with Tailwind CSS featuring bold shadows, thick borders, and emerald accents
Responsive layout supporting mobile through desktop
Real-time loading states with contextual flavor text ("Analyzing requirements...", "Breaking down architecture...")

Backend (NestJS + TypeScript)

RESTful API architecture with modular service design
Google Gemini AI integration for intelligent roadmap generation
Structured prompt engineering to ensure consistent, parseable roadmap output
Google Calendar API integration for availability checking and event creation

Authentication and Data

Supabase for authentication (Google OAuth) and PostgreSQL database
Secure session management with HTTP-only cookies
User profiles storing calendar connection status and preferences

Key Technical Decisions

Chose NestJS for its modular architecture and TypeScript-first approach
Used Gemini AI for its strong structured output capabilities and cost efficiency
Implemented neo-brutalism UI for a distinctive, memorable developer experience
Server Components in Next.js for faster initial page loads

Challenges we ran into

1. OAuth Token Flow Complexity

Google OAuth with Supabase presented unexpected challenges. The access token arrives in the URL hash fragment, which is not sent to the server. We had to implement a client-side handler to extract tokens from the hash and establish the session before redirecting to the app.

2. Structured AI Output Parsing

Getting Gemini to consistently return valid JSON roadmaps required extensive prompt engineering. We implemented strict output schemas and fallback parsing to handle edge cases where the AI would include markdown formatting or explanatory text around the JSON.

3. Calendar Availability Logic

Finding open slots in a user's calendar while respecting working hours, existing events, and timezone differences proved more complex than anticipated. We built a custom algorithm that:

Fetches events for a configurable time window
Identifies gaps between events
Filters by preferred working hours
Suggests optimal scheduling based on task duration

4. Next.js 16 Async API Changes

Next.js 16 introduced breaking changes where cookies() and headers() became async. This required refactoring our authentication checks across all server components to properly await these functions.

Accomplishments that we are proud of

End-to-end execution flow: From a single sentence to calendar events in under 30 seconds
Intuitive planning depth modes: Users can choose their level of detail without complex configuration
Distinctive UI/UX: The neo-brutalism design makes Aedar instantly recognizable and enjoyable to use
Production-ready architecture: Clean separation of concerns, comprehensive error handling, and type safety throughout
Calendar integration that actually works: Real availability detection and event creation, not just a mock feature

What we learned

Technical Lessons

Prompt engineering is as important as traditional engineering when building AI products
OAuth flows have many edge cases -- always test with fresh accounts and incognito windows
Neo-brutalism CSS is surprisingly complex to implement consistently (those shadow offsets add up)

Product Lessons

Developers want tools that reduce friction, not add new workflows to learn
The "human-in-the-loop" confirmation step before auto-scheduling builds trust
Loading states are a UX opportunity, not just a necessary evil

Process Lessons

Building the happy path first and handling edge cases iteratively kept momentum high
TypeScript's strict mode caught numerous bugs before they reached production
Testing with real calendar data (our own schedules) revealed issues synthetic data never would