Inspiration
Friends share small things all the time, like wanting to learn photography or visit a new coffee shop, but those details often get buried. We built Constellation to help people remember these moments and turn them into real plans.
What it does
Constellation lets users save memories about their interests, goals, availability, and things they can share. Users can join shared circles with friends or create a personal circle for people who do not have accounts. The app finds meaningful connections between memories and suggests an activity that always includes the signed-in user.
Our Chrome extension brings this experience to Instagram and Facebook. When a user chooses a post, it compares the caption with their memories and suggests a related activity they could do with a friend. They can save the idea or write an invitation without leaving the page.
How we built it
We built Constellation with Next.js, React, SQLite, React Flow, and a Chrome extension. Meta’s Muse Spark turns natural writing into structured memories and reasons across a circle’s memories to create practical activity suggestions. Each suggestion includes the exact memories supporting it, which we display as an interactive path in the constellation map.
Backboard provides long-term memory. We create a separate memory index for each person in a circle and store only approved information. During discovery, Backboard semantically recalls relevant memories, including older details, before Muse creates a plan. We validate every returned person and memory ID so the AI cannot introduce unsupported information.
Users review memories before saving them and control whether each memory is shared, summarized, or private. Private memories are excluded from group discovery.
Challenges we ran into
Reading Instagram posts was surprisingly difficult because captions, locations, comments, and reply buttons often use similar page structures. We built extraction rules to identify the original caption and added a review field in case Instagram changes its layout.
We also spent a lot of time making the recommendations trustworthy. The system must use real memories, include the signed-in user, respect privacy, and avoid inventing someone’s availability or skills. Structured AI responses and server-side validation helped us solve this.
Accomplishments that we're proud of
We are proud that Constellation uses AI for more than generating text. It remembers what people care about, finds a real opportunity between friends, explains its reasoning, and helps someone take the next step by writing an invitation.
We are especially proud of the interactive “Why this connection?” view and the browser extension, which turns something discovered while scrolling into a possible real-world plan.
What we learned
We learned that AI becomes much more useful when it combines reasoning with long-term memory. Muse understands what people share and finds meaningful combinations, while Backboard makes older memories useful at the right moment.
We also learned that user control is essential. Letting people review memories, choose privacy settings, inspect the evidence, and decide whether to send an invitation makes the experience feel more personal and trustworthy.
What's next for Constellation
Next, we want to deploy shared circles across devices, improve authentication, and notify users when a newly added memory creates a promising connection. We also want to learn from which suggestions become real plans, helping Constellation make increasingly useful recommendations while keeping users in control.
Log in or sign up for Devpost to join the conversation.