Persona: AI Market Research

🕵️ Effortless Market Research

Conducting first-party research can be difficult, time-consuming, and expensive. As a computer science and business student fascinated by startups, I’ve constantly seen how inaccessible real, first-party consumer insights are for early-stage founders. That's why I decided to build Persona, a web app where gathering high-quality audience insights is as simple as asking a question.

🦾 What It Does

AI-Powered Customer Generator: Over 100 unique potential customers are created in under a minute with diverse demographics and backgrounds.

Emulated Customer Profiles: Agents are then dispatched to impersonate customer profiles and emulate them to provide deep market insights

Survey Response Breakdown: Survey results can be broken down statistically, including bar charts, averages, and more

Individual Response Breakdown: View the individual responses of Emulated Personas and evaluate which customer segment they would fall under

3D Geo-Spatial Visualization: Visualize where your potential customers are around the world

🛠️ How Its Built

See the attached diagram for more details.

Frontend

React Router v7
TailwindCSS
JavaScript/TypeScript
Vite
Lucide Icons

Backend

Node.JS
Cloudflare Pages
Cloudflare Workers

Agentic GenAI

Google Gemini 2.5 Flash
OpenRouter
n8n Workflows

⛔ Challenges

n8n was used to easily automate the backend that would interface with LLM providers. Its a relatively straightforward visual coding tool but this introduced complexity. The lack of shared LLM memory across concurrent threads meant that personas had to be generated sequentially with careful cross-run context, otherwise the model repeated the same persona over and over again. On the UI side, building a 3D globe was far from trivial. My limited experience with Three.js and shader work made atmospheric effects, point mapping, and interaction design much harder than anticipated. Additionally, since this was a solo project, every bug, every design choice, every technical misstep landed on me alone.

🧠 Accomplishments & Growth

Despite the technical hurdles, the 3D interactive globe looks awesome. I wanted to implement an atmosphere haze and a Karman line glow that would immerse you in the data. I successfully built the core experience: generating personas, collecting survey responses, and visually exploring results in a single interface. It was my first project connecting a UI to an n8n-powered backend workflow, and the initial integration came together shockingly fast. I've used GitHub Copilot for years but I had yet to try Agents until this weekend. This was also my first hackathon where I went well past the 12-hour mark!

🤔 What's Next?

You're nearing the end of this so I'll make it quick:

Interactive Emulated Personas: Ask follow up questions or reasoning behind responses
Sentiment Analysis and Summarization: Analyze short and long open ended questions
More Visualizations: Demographic breakdowns in the form of pie charts
Cleaner Code: I wanted to move fast and iteratively design my prototype, so there's a lot to be cleaned up 😅
No more n8n: While nice and simple, its too simple for this use case and requires code