Inspiration Whether it's a robust backend to some software system or a distributed cloud app, every engineering masterpiece begins with a vague human notion. Yet, there lies a huge gap between free-form AI dumps and structured development. Thousands of wasted developer hours are spent on converting their ideas to a structured schema, tracking task dependencies, and estimating timelines before they even write their first line of code.
That's why I created Breakdown, an engineering tool capable of removing the cognitive overhead and delivering structured product concepts instantly from raw thoughts.
What it does Breakdown is a sleek, dark theme-based workspace that compiles raw directives into a structured Compiled Architecture Matrix.
Users can type or dictate the goals of their project using a built-in voice module.
The system immediately distills ideas into a top-level Item, analyzing them through automated algorithms and generating a Feasibility Score and Realism Metric.
It divides projects into granular, prioritized Steps, while automatically outlining estimated timelines and target implementation complexities.
With a single click, it creates an interactive dependency graph that visually displays the exact structure and order of tasks that need to be completed first.
How I built it I developed the application from scratch using a lightweight and fast stack:
Frontend: Standard HTML5, adaptive CSS variables, and native JavaScript. I used the Web Speech API for voice dictation and Cytoscape.js to dynamically render and manage node dependency graphs.
Backend: A Node.js and Express-based architecture handling structured API endpoints.
Database and AI Layer: I integrated the Gemini API for advanced parsing and structural analysis. Under normal conditions, session history logs are managed through Prisma ORM connected to a cloud PostgreSQL relational database.
Challenges I ran into One of our biggest obstacles was dealing with unstable cloud database connections during development. Poor network conditions often caused connection timeouts, which could freeze or crash the entire application pipeline.
To solve this, I built an asynchronous try-catch container on the backend. If the Prisma client loses its connection handshake, the server silently handles the issue without crashing, bypasses the database entirely, avoids returning a 500 error, and switches into a sandbox transmission mode where structured AI data is delivered directly back to the frontend.
Another issue I ran into
Accomplishments that we're proud of Database Fallback Mode: I created a complete fallback state that allows the application to continue functioning even without cloud database access.
Scannable UI/UX: I designed a highly focused, distraction-free workspace that feels fast, technical, and lightweight.
Relational Algorithmic Mapping: I engineered a pipeline that converts unstructured text into matching relational arrays across both our database schemas and graph visualization systems.
What I learned Throughout development, I learned the importance of architectural resilience. Building the database fallback system taught us how critical graceful error handling is when cloud services become unreliable. I also gained valuable experience with prompt engineering constraints to ensure the AI consistently returns clean, structured JSON instead of unpredictable conversational responses.
What's next for // BreakDown I plan to evolve Breakdown from a project mapping workspace into a full code scaffold environment. Our next major feature is automated export functionality. Soon, users will be able to click any step within their architecture graph and instantly generate downloadable Prisma schemas, starter Express route files, and Docker configurations directly from their mapped project structure. Another thing that I would want to do is fix some of the issues in the product as of right now. Evidently, storing conversations doesn't work, and the Graph explosion portion doesn't work, so being able to put both of these together would help make the project more put-together.
Another expansion of this project would be to allow for file uploads, specifically, images. This way, if Breakdown saw a visual product, it could try to break that down and the steps it took to make that process.
AI USED IN THE PROCESS:
Gemini 3.5 Flash was used to do 2 key things. 1: Generate a color and font scheme to help me make a modern and sleek UI for the website
2: Used to make the prompt to feed into the Gemini API Model to make Breakdown OS
SUBMITTING THE DEMO AND VIDEOS BY EMAIL. UPLOADED YOUTUBE VIDEO IS A RANDOM VIDEO
Log in or sign up for Devpost to join the conversation.