The problem Cognivo solves

Modern knowledge work is fragmented. Ideas, goals, tasks, and plans live across browser tabs, notes apps, and scattered documents — with no spatial structure and no way to see how they connect. Traditional tools are linear and passive; they store information but don't help you think or plan.

Cognivo is an AI-powered infinite visual canvas for productivity and life-mapping. It turns scattered thoughts into structured, interconnected knowledge — and actively helps you work smarter:

1. Smart Brain Dump: Speak or paste your thoughts; AI automatically sorts them into the Eisenhower Matrix (Urgent/Important grid) so you instantly know what to act on.
2. YouTube Roadmap Extractor: Drop any YouTube URL and get a step-by-step learning roadmap extracted directly onto your canvas — no manual note-taking.
3. Conflict Detection: Flags overlapping deadlines and illogical task dependencies before they derail your plans.
4. Time Reality Check: Warns you if your total workload exceeds available time before a deadline — a reality check your to-do list never gave you.
5. Persistent Canvas: Everything auto-saves to localStorage. Your second brain is always where you left it.
6. Voice Input: Speak thoughts directly onto the canvas via Speech-to-Text (Chrome/Edge).

Whether you're a student mapping a semester, a founder planning a product, or anyone trying to turn mental chaos into clarity — Cognivo makes thinking visible and planning honest.

Challenges we ran into

1. Persistent canvas state across sessions: React Flow's node/edge state doesn't persist by default. We implemented Zustand for global state management and wired it to localStorage, but had to handle edge cases where re-hydrating large canvas states caused node position drift and z-index conflicts.

2. Gemini API structured output: Getting Google Gemini 1.5 Flash to return clean JSON (for Eisenhower classification and conflict detection) reliably required careful prompt engineering. Early responses mixed JSON with markdown fencing, breaking our parsers — fixed with strict output format instructions and fallback sanitization.

3. YouTube transcript extraction: The YouTube Roadmap Extractor relied on transcript availability, which isn't guaranteed for all videos. We handled missing/auto-generated transcripts gracefully and tuned the Gemini prompt to generate a coherent roadmap even from sparse input.

4. Time Reality Check logic: Calculating whether a user's total task workload exceeds available hours before a deadline required normalizing across different node types, incomplete deadline fields, and missing priority data — all entered freeform by the user.

5. Render cold start latency: Deployed on Render's free tier, the app had ~30–50 second cold start times that made demos feel broken. We added a loading state and a "warming up" indicator to manage user expectation during cold boots.

6. Voice input cross-browser compatibility: The Web Speech API behaves differently across Chrome, Firefox, and Safari. We scoped support to Chromium browsers and added a clear fallback message for unsupported environments.