💡 Inspiration
We all struggle with information overload. Every day, we generate messy meeting notes, download endless PDFs, and dump ideas into chat apps. But raw data isn't useful until it's organized.
We asked ourselves: What if we could have a "Second Brain" that doesn't just store notes but actually understands them?
That led to Gemini Synapse. We wanted to build a bridge between chaotic human thought and structured project management, empowering users to turn a brain dump into a project board in seconds.
🧠 What it does
Gemini Synapse is an intelligent productivity assistant.
- Input: Users paste raw meeting notes, brain dumps, or upload PDF documents.
- Process: The system uses Google's Gemini 1.5 Pro to analyze the context, identify key entities, and separate signal from noise.
- Output: It generates a structured Dashboard containing:
- Projects: Active initiatives with status and due dates.
- Tasks: Actionable items assigned to specific people.
- Decisions: Key choices made, along with their impact level.
⚙️ How we built it
We focused on organizing and retrieving professional knowledge from unstructured content. The key capability is Structured Outputs: we constrain Gemini to return a strict JSON object (Projects, People, Decisions, Tasks, and Summary). This structured JSON is central to the product because it directly powers the UI—turning messy notes and PDFs into reliable project cards, editable task lists, and decision logs instead of a fragile, free-form chat response.
To balance speed and depth, we tune Gemini's thinking level: we run lightweight extraction for fast organization and switch to deeper reasoning when generating cross-note summaries or prioritizing tasks.
For PDF inputs, Gemini’s document understanding extracts key facts, decisions, and action items from long-form content and converts them into the same structured schema, enabling consistent recall across different sources.
Tech Stack:
- AI Model: Gemini 1.5 Pro (via Google AI Studio)
- Frontend: React, TypeScript, Tailwind CSS
- Integration: Gemini API for multimodal (text + PDF) processing
🚧 Challenges we ran into
- Taming the LLM: Getting the AI to consistently output valid JSON for edge cases (like vague meeting notes) was difficult. We had to iterate heavily on our system instructions and schema definitions.
- Nuance Detection: Teaching the model to distinguish between a "suggestion" and a firm "decision" required fine-tuning our prompts with few-shot examples.
- PDF Complexity: Parsing multi-column PDFs and retaining the correct context for tasks was a challenge that Gemini's long-context window helped solve.
🏆 Accomplishments that we're proud of
- Zero Hallucinations in Structure: We successfully constrained the model to strictly follow our JSON schema, ensuring the dashboard never breaks.
- The "Magic" Moment: Seeing a messy, 3-page PDF instantly convert into a clean, actionable Kanban board for the first time was incredible.
- Seamless UI: We built a "Modern Classic" aesthetic (Ivory/Cream gradients) that makes the tool feel professional yet welcoming.
🚀 What's next for Gemini Synapse
- Direct Integrations: Connecting our structured JSON output directly to tools like Notion, Asana, and Jira.
- Voice Mode: Adding real-time voice transcription to organize meetings as they happen.
- Multi-file Synthesis: Allowing users to upload 10+ PDFs and ask questions across the entire knowledge base ("What did we decide about the budget across these 5 meetings?").
Built With
- ai
- api
- gemini
- javascript
- node.js
- python
- studio
- tailwin
- typescript
Log in or sign up for Devpost to join the conversation.