# Cognos: Full Project Story
## Inspiration
We drew inspiration from mind mapping and whiteboarding—the way ideas naturally form when visualized. As UX designers, we've experienced how effective diagramming is: jotting down thoughts spatially reveals connections that linear formats hide. We wondered: what if AI conversations could work the same way?
## What it does
Cognos shifts from chat-centric to object-centric cognition. You're not managing conversations—you're grappling with topics that evolve, questions that clarify, ideas that connect across sessions in non-linear ways. Cognos makes this visible.
- **Object Creation:** Analyzes your chat history and categorizes conversations into "Objects"—the main topics and anchor points of your thinking.
- **Node Exploration:** Click into any Object to see its "Nodes"—smaller units like milestones, pivot points, facts, or concepts that form the structure of your thoughts.
- **Flexible Interaction:** View relationships between nodes at a glance, or dive deep into specific ones. Add temporary notes, discuss the entire Object with AI, or go deeper on a particular Node.
- **Dynamic Growth:** As you continue conversations, new connections and nodes emerge—just like how our minds evolve.
- **Smart Merging:** Start a new topic directly in Object view. AI identifies if it relates to existing Objects or creates a new one for fresh exploration.
## How we built it
As two UX designers without coding backgrounds, we leveraged Gemini AI throughout the entire development process—treating it as a team member who understands design, engineering, and more. We followed our UX process: defining user flows, creating screens, and iterating—but with AI as our development partner.
**Tech Stack:**
- **Design to Code:** Imported Figma designs directly into Google AI Studio
- **Deployment:** Google Cloud Run, connected via project API
- **AI Features:** Leveraged Google AI Studio's built-in capabilities:
- **Gemini Intelligence:** Core reasoning and analysis engine
- **Google Search:** Enhanced context and fact-checking
- **Thinking Mode:** Deeper cognitive processing for complex topics
- **Fast Response:** Optimized user experience during topic discussions, chat history analysis, and categorization
## Challenges we ran into
- **Prompt Engineering Gap:** Despite AI tools being "natural language," there's a significant difference between how designers think and how AI interprets instructions. We spent considerable time learning to communicate our design intent clearly to Gemini.
- **Processing Speed:** Analyzing chat history accurately and quickly proved difficult. Even after integrating Gemini, performance remained semi-ideal due to our limited knowledge of system optimization.
- **Chain-Effect Debugging:** Vibe coding adjustments often triggered cascading bugs. Without understanding backend logic, we sometimes hit dead ends when trying to fix issues, even when providing Gemini with error messages and clear objectives.
- **Format Decision:** We grappled with whether Cognos should be a plugin or web app. Plugins could only read current chats (missing the full picture), while web apps faced privacy concerns around accessing chat history. We ultimately created a web app with a chat history upload feature—mimicking the experience of AI analyzing previous conversations while respecting privacy.
- **File Size Limitations:** The system struggled with large chat history files, likely due to our limited optimization knowledge.
## Accomplishments we're proud of
As UX designers comfortable with user flows and screens, shifting to hands-on building was a major accomplishment. We successfully integrated AI into our entire iteration process—similar to our design process, but with Gemini as an additional team member.
We built and iterated something from scratch following UX principles, demonstrating and testing an idea we're both passionate about.
## What we learned
- **Learn by Doing:** This project exemplified learning through action. We started with limited understanding of vibe coding but dove in anyway—putting ideas down and involving AI throughout. We learned from obstacles, failures, and restarts. That's the mindset needed in the AI era.
- **Use AI to Talk to AI:** We discovered prompt engineering itself can be AI-assisted. When stuck on how to communicate technical iterations, we asked multiple AIs how to better prompt them. This helped us fix bugs and understand what AI comprehends.
## What's next for Cognos
We're committed to continued refinement:
- **Multi-Layer Object Map:** Beyond listing Objects, show relationship strength and connection types between them.
- **Nested Node Views:** Enable drill-down into multi-layered nodes as topics deepen, revealing increasingly granular thinking structures.
- **Timeline View:** Show how knowledge and cognition evolve over time, helping users understand where they are in their thinking process.
- **Plugin Development:** Build and launch a plugin for real-world testing and user feedback.
As UX designers, we're excited to keep iterating based on how people actually use Cognos in their thinking processes.
Log in or sign up for Devpost to join the conversation.