Inspiration

The core inspiration for our project was to develop a tool that could effectively solve the practical challenge of keeping track of daily tasks and managing complex information quickly, a critical need for busy individuals like university students. Our team aimed to build a rapid, high-quality prototype of a task/collaboration system. To achieve this ambitious goal within the hackathon time limit, we developed a novel methodology: "Pure Vibe Coding," centered around the Kiro development agent.

What it does

The primary functional component of the resulting prototype was a Real-Time Collaboration Engine, designed for our main application (adiutor/App.tsx). This engine is essential for a dynamic task management tool, as it ensures all users stay synchronized instantly. Key features of this component included:

  • A real-time data service setup for live data synchronization, integrated with authentication and safety checks.
  • A debounce system to optimize performance by preventing excessive writes to the custom backend during rapid user input.
  • A responsive UI built with Tailwind grid layouts, ensuring the tool is easily usable across mobile devices and the Electron desktop application (managed by adiutor/electron/main.cjs).

How we built it

The project was built using a highly efficient, agent-driven methodology based on conversational flow and automation, specifically targeting files within the Durween/adiutor/ structure:

Vibe Coding Methodology

We treated interaction with Kiro like a pair-programming session with a senior developer. Instead of detailed specifications, we used short, conversational prompts to describe the desired behavior (e.g., how the StatsBoard.tsx component should update) and iteratively refined the generated code immediately. The rapid changes were reflected across core files like App.tsx and the primary entry point, index.tsx.

Automated Workflows

  • Agent Hooks: We implemented hooks to automate repetitive setup and housekeeping. One hook applied linting and formatting automatically upon code generation, referencing rules in tsconfig.json and package.json, while another prepared necessary dependencies and file layouts (e.g., generating initial structure for components like AduitorCharacter.tsx).
  • Steering Documents: These served as a project-wide style guide to ensure all generated code was consistent. This mandated using only Tailwind CSS (for full responsiveness) and adopting the style of a "cautious and pragmatic senior engineer."

Multi-Capable Platform (MCP)

MCP enhanced Kiro's capabilities, making it a better full-stack collaborator by enabling:

  • Web browsing to pull in the latest documentation for modern libraries, helping Kiro generate code for specific services like the geminiService.ts.
  • Generating code with a structured, multi-directory mindset, keeping the file hierarchy clean, even though the hackathon code was contained in a single repository.

Challenges we ran into

The main challenge was discovering that a more structured, traditional approach to development proved too slow for the hackathon environment:

  • Spec-Driven Bottleneck: An initial attempt at spec-driven development required creating detailed Markdown explanations for the data model. Once iteration began, small design changes necessitated updating large sections of the spec, which significantly slowed down momentum and was quickly abandoned.

Accomplishments that we're proud of

The team’s greatest source of pride was demonstrating Kiro’s capacity to handle complex, large-scale requirements in a single request:

  • We successfully prompted Kiro to build the entire 400-line Real-Time Collaboration Engine component in one response, directly impacting files like App.tsx. This complete component was fully functional, combining complex elements like custom backend synchronization, debouncing, and responsive UI design.

What we learned

The key takeaway was the confirmation of the "Vibe Coding" approach's effectiveness for high-speed, high-quality development:

  • Flexibility and Speed: We learned that Vibe Coding is faster and far more flexible than traditional spec-driven development for quick iteration. By adapting to immediate, one-sentence changes, it leveraged Kiro's short-term context to keep the generated code accurate and consistent without the overhead of spec maintenance.

What's next for Durween

To solidify its value as a next-generation productivity tool for students, Durween could evolve by integrating social accountability and focus mechanisms directly into the real-time collaboration engine.

Social Accountability and Focus Features

  • Study Session Co-Working: Implement a feature where users can initiate shared "focus sessions" (e.g., using the Pomodoro Technique) with other Adiutor users. Success could be gamified by tracking group focus time or having a shared reward, mirroring the concepts in popular focus apps like Forest or Flora.
  • Distraction Guard: Utilize the real-time engine to build an optional, temporary "Distraction Guard" function. When a focus session is active, this feature could monitor for and notify the user about attempts to open common distracting apps (like social media platforms). This integrates directly with the stated goal of solving the challenge of rapid information management by removing procrastination barriers.

Agent Enhancement

  • AI Contextual Checking: Leveraging Kiro's MCP, the tool could be enhanced to proactively perform a "Social Media Wellness Check." For instance, if a major deadline is looming in the user's task list, the tool could gently recommend a temporary block on a chosen social media app for the next two hours, helping the student prioritize based on their own data.
  • Group Project Facilitation: Expand the collaboration engine to offer direct, streamlined integration for group projects. This would include features like automatically assigning tasks to Adiutor users, viewing individual progress dashboards, and coordinating scheduled group study times, building on the multi-user capabilities already demonstrated by the Real-Time Collaboration Engine.

Built With

Share this project:

Updates