Synapse

Inspiration

We write down our thoughts, but rarely connect the dots between them. A worry from January might become an accomplishment in October, buried hundreds of entries later.

We imagined a “Spotify Wrapped” for journaling: a way to see recurring thoughts, evolving interests, and personal growth. Inspired by connections between neurons, we designed Synapse to turn individual memories into an explorable map.

What it does

Synapse is a multimedia journaling app where users can type, record voice memos, photograph handwritten pages, or chat with AI to create an entry.

Topics become neurons in a knowledge graph. Their brightness reflects how often they appear, while connections represent topics mentioned together. Users can explore their entire journal or compare monthly and yearly views.

An “Ask your journal” feature supports open-ended questions about past experiences. Check-ups revisit recurring intentions, while Wrapped reports highlight themes, changing perspectives, and milestones with links to supporting entries.

How we built it

We divided development into backend and database, ML and knowledge graph processing, and frontend visualization, then collaborated on insights and Wrapped.

Our pipeline converts multimedia inputs into editable text before analysis. Structured extraction identifies topics, people, intentions, and supporting passages. Embeddings help resolve differently worded references to the same topic.

Graph calculations count distinct entries and topic co-occurrences. For open-ended questions, retrieval-augmented generation finds relevant passages and supplies them to the language model, grounding answers in the user’s journal.

Challenges we ran into

Personal writing is messy. The same idea can appear under different names, handwriting can be difficult to read, and voice recordings can contain transcription errors.

Another challenge was interpreting change responsibly. Mentioning someone less often doesn’t prove a relationship weakened, and writing about a goal doesn’t mean it was achieved. We designed around reviewable text, supporting evidence, and user confirmation.

Keeping the graph stable across time periods was also important: users should notice changes in their thoughts without having to relearn the layout.

Accomplishments that we're proud of

We’re proud of bringing several ways of journaling into one coherent experience and making the journal itself the foundation for personalized answers.

We’re especially proud of the neuron-map concept: users can move from a broad pattern to the specific memories behind it. Connecting that exploration with reflection and goal tracking makes the visualization useful beyond its appearance.

What we learned

We learned that meaningful personalization depends on how information is organized and retrieved, not just how well a model writes.

We also learned to separate measurable patterns from interpretations. Topic counts can show what occupies someone’s writing, but the person remains the authority on what those patterns mean.

What's next for Synapse

We want to improve retrieval across years of entries, make topic corrections easier, and develop richer timelines connecting intentions with later milestones.

We also want to refine multimedia capture and learn which check-ups users actually find helpful, so Synapse becomes a journal people return to for both recording and reflection.

Built With

• gemini
• postgresql
• python
• supabase
• typescript