AIdeaText: Disccovery conText

Introducing AIdeaText: Discovery conText, the multimodal research partner that turns the "noise" of discovery into a structured map of knowledge.

Comment

Inspiration

The inspiration for AIdeaText: Discovery conText stems from the fragmented nature of modern research. Researchers, students, and creators often find themselves overwhelmed by "information silos"—voice memos, video lectures, and long PDF documents that remain disconnected. We wanted to build a tool that doesn't just store data but actively helps the human brain connect the dots in real-time, turning the chaotic process of discovery into a structured, visual map of knowledge.

What it does

AIdeaText is a multimodal research partner that bridges the gap between raw input and structured insight. It allows users to:

  • Interact in Real-Time: Talk to a Gemini-powered conversational agent that listens and brainstorms with you.
  • Visualize Knowledge: Watch as a dynamic 3D knowledge graph automatically populates with entities (people, places, concepts, dates) extracted from your conversation.
  • Process Multimodal Media: Upload text, audio, or video files to extract hidden relationships and add them to your research map.
  • Synthesize Arguments: At the end of every session, the AI performs a deep analysis to identify logical links between concepts, providing a "second brain" that reveals the structure of your arguments.
  • Persistent Research: Save and revisit past sessions to continue building complex research projects over time.

How we built it

We built AIdeaText using a high-performance full-stack architecture:

  • Frontend: React 19 and Vite for a lightning-fast UI, styled with Tailwind CSS for a professional "Technical Dashboard" aesthetic.
  • Intelligence: Integrated the Gemini 2.5 Flash and Gemini 3.1 Pro models via the @google/genai SDK. We utilized the Live API for low-latency voice interaction and real-time multimodal processing.
  • Visualization: A custom-built interactive graph engine using D3.js to handle complex relational data.
  • Backend: An Express server managing persistent storage (JSON-based DB) and OAuth2 flows for secure Google Drive integration.
  • Animation: Framer Motion (motion/react) for smooth transitions and an immersive user experience.

Challenges we ran into

One of the primary hurdles was managing the real-time synchronization between streaming audio and the visual graph. Ensuring the AI could identify entities and update the UI without interrupting the flow of conversation required optimized state management. Additionally, implementing OAuth2 within an iframe environment presented significant cross-origin challenges, which we overcame by designing a robust popup-based communication system using postMessage.

Accomplishments that we're proud of

We are incredibly proud of the Automated Relationship Extraction feature. Seeing the AI analyze a 20-minute conversation and instantly draw logical connections between scattered ideas feels like a breakthrough in research productivity. We also successfully implemented a Multimodal Processing Pipeline that treats text, audio, and video as equal sources of truth, allowing for a truly holistic research experience.

What we learned

This project deepened our expertise in Real-Time AI Orchestration. We learned how to balance the latency requirements of voice interaction with the computational cost of deep entity extraction. We also gained significant insights into Human-AI Collaboration, specifically how an AI agent should transition from a passive listener to an active brainstorming partner to best support the creative process.

What's next for AIdeaText: Discovery conText

The roadmap for AIdeaText includes:

  • Web Grounding: Integrating Google Search to allow the agent to verify facts and cite sources in real-time.
  • Collaborative Mapping: Multi-user rooms where teams can build a shared knowledge graph simultaneously.
  • Export to Manuscript: A feature to instantly convert a research graph into a structured Markdown outline or a first draft of a paper.
  • Deeper Drive Integration: The ability to "index" entire Google Drive folders to find connections across years of personal research.## Inspiration

Built With

Share this project:

Updates