Concept Compass

Inspiration

Every great project starts with a single idea, but the path from that initial spark to a full concept is often blocked by the dreaded blank page. I've always felt that traditional brainstorming can be too linear, limited by my own knowledge. I wanted to build a thinking partner that wouldn't just give me answers, but would help me ask better questions and discover paths I hadn't even considered.

What it does

Concept Compass is an interactive web app that turns any keyword into a dynamic, explorable mind map. Instead of a static list, it generates a visual graph of interconnected ideas that you can navigate infinitely. The process is simple:

1. Plant a Seed: You start with a single concept.
2. Grow the Universe: The AI engine builds a rich map of related concepts around it.
3. Explore Infinitely: This is the core of the app. You can click on any node to make it the new center, instantly generating a fresh set of connections and allowing for an incredibly deep exploration of any topic.

How I built it

I built Concept Compass in a 48-hour sprint using a spec-driven development workflow, with Kiro.dev acting as an AI co-pilot. This allowed me to define the project's architecture and requirements upfront, which then guided the AI in generating a great, high-quality starting point for the codebase.

The tech stack is modern and high-performance:

• Framework: Next.js 15 (App Router & Turbopack)
• Language: React 19 & TypeScript
• Styling: Tailwind CSS & shadcn/ui
• AI Models: OpenAI's gpt-oss-20b (development) & gpt-oss-120b (production) via the OpenRouter.ai API
• Visualization: React Flow
• Testing: Vitest & React Testing Library

The backend is a resilient Next.js API route that I engineered with timeouts and robust error handling to ensure a smooth user experience, even if the AI service is slow or returns unexpected data.

Challenges I ran into

The main challenge was the inconsistency of the AI's output. I expected a clean JSON array of strings, but the gpt-oss model would occasionally return malformed data or time out. I solved this by engineering a robust frontend state machine that gracefully handles these errors, showing a clear message and allowing the user to retry the request.

Another challenge was the visual design. Getting the mind map to look and feel "alive" and intuitive was a process of trial and error. Given the tight deadline, I focused on creating a clean, functional, and visually appealing interface, reaching a solid MVP that I can continue to build upon.

Accomplishments that I'm proud of

As a solo developer, I'm incredibly proud of building a functional, stable, and polished full-stack AI application in such a short time.

Specifically, I'm proud of engineering a highly resilient application that gracefully handles API errors. It's one thing to make an app that works on the "happy path," but building in timeouts, retries, and clear error states makes it feel like a real, production-quality product.

What I learned

This project was a deep dive into the practical application of gpt-oss models. My main takeaway is how powerful these models are for creative and divergent thinking. Their ability to generate not just logical but also surprising, "out-of-the-box" connections is the true engine of Concept Compass.

I also learned how effective a hybrid human-AI development workflow can be. Using an AI assistant for the heavy lifting of scaffolding and initial code generation allowed me to focus my time on what mattered most: the core user experience, the visual polish, and the complex task of prompt engineering.

What's next for Concept Compass

This MVP proves the core concept, but my vision for Concept Compass is much larger. The next steps are focused on transforming it from a simple tool into a comprehensive creative suite:

• Branching History & Exploration Trees: I plan to implement a system to go back and forth in the graph of ideas. This would allow a user to explore multiple creative paths from a single starting point and visually compare the different branches of their thought process.
• Persistent Brainstorming Sessions: A crucial next step is to save entire exploration sessions to a user's account. This would allow for long-term projects, enabling users to pause and resume their creative process at any time, without losing their history.
• "Deep Dive" Elaboration Engine: To make the tool more useful for research and learning, I'll add a feature to expand the content of any node. Clicking an "Elaborate" button would trigger a new AI call to generate a detailed summary, key points, or an explanation of that specific concept.
• Actionable Content Generation: The ultimate goal is to turn ideas into action. I envision a feature where a user can select a path of nodes and have the AI generate practical content from it, such as a blog post outline, a series of social media posts, or key talking points for a presentation.

Built With

• gpt-oss
• kiro.dev
• nextjs
• openai
• openrouter
• react
• reactflow
• shadcn
• tailwindcss
• vercel
• vitest