StarsHD

Inspiration

The idea started with wanting to use something as vast and universal as stars to help with mental health, specifically ADHD. Early on, one thought stuck: what if every star a person could see was tracked and meant something, could you actually know about every star in the world? Very quickly it became clear that's an impossible task, there are too many stars, too much information, no human mind can hold or track all of it. That realization ended up being the actual insight behind the project. A brain trying to hold every star in its head is exactly what an ADHD brain trying to hold every task in its head feels like, too much to track, no natural place to put it. Instead of trying to solve the impossible problem of tracking every star, the project became about using a real, mapped subset of stars as a place to put something a person actually can track: their tasks. The sky became the external memory, not the mind.

What it does

StarsHD turns each task into a real star in a sky built from real astronomical data. Brightness reflects urgency, color reflects how much effort or energy a task takes, and related tasks cluster into constellations instead of sitting in a flat list. Instead of asking the user to sort and prioritize everything themselves, the app does that work: it takes in everything on their mind at once and highlights a single starting point, removing the moment where ADHD task-switching usually breaks down. Completed tasks don't disappear, they stay lit permanently, building a personal sky over time instead of an inbox that's always chasing zero.

How we built it

The project was built using Claude for the core reasoning and content generation, LangChain to structure and manage the AI logic, and API integrations to connect everything together. Claude handles interpreting a user's unsorted brain dump and identifying what to surface first, LangChain manages the flow and chaining of those AI steps, and the API layer connects the AI logic to the real astronomical data and the app itself.

Challenges we ran into

Development itself was a real challenge, getting the different pieces (the AI logic, the data layer, the interface) to work together smoothly took a lot of iteration. Finding the right data was just as hard. Real astronomical data exists, but pulling it into a form that could actually map cleanly onto tasks (brightness to urgency, color to effort, distance to time) took real trial and error to get right rather than just decorative.

Accomplishments that we're proud of

We're proud of the product we ended up with, and honestly we learned a lot along the way. Working with Claude and LangChain together taught us a lot about structuring AI logic properly instead of relying on a single prompt to do everything, and working with real data taught us a lot about the gap between having data and having data that's actually usable in a product.

What we learned

We learned a lot about building with AI in a structured way, connecting language models to real data, and about the ADHD-specific problem itself, that the real barrier is often not motivation but holding too much in mind at once. We also learned a lot about how much iteration real data integration takes compared to how simple it looks at the start.

What's next for StarsHD

Next steps include verifying the approach with health officials and professionals who work directly with ADHD, to make sure the mechanisms in the app are genuinely helpful and not just well-intentioned. Alongside that, we want to keep refining the data integration, expand testing with real users, and keep improving how the AI decides what to surface first as the product grows.

Built With

• claude
• deepseek
• langchain
• react-native