About the Project

Inspiration

“You should be using AI not like Ozempic, but like a detailed workout plan.” In other words, AI should be used as a collaborative tool, not as something that does the work for you. These words, spoken by Toptal Chief Customer Officer of AI Jeff Mills at the fireside chat this morning, inspired us to create AskBetter, an AI prompt analyzer that evaluates whether or not you are using AI in a smart and innovative way. Assessing users on Autonomy, Curiosity, Critical Thinking, and Engagement, AskBetter holds up a mirror to show people what kinds of questions they are asking, where they are being passive, and how to ask better. The name says it all: better questions, better answers.

What it does

AskBetter analyzes your ChatGPT, Gemini, or Perplexity conversations and shows you how you're actually using AI. It classifies every prompt you send by intent — delegation, curiosity, collaborative, or verification — and scores your conversation across six quality dimensions. You get a breakdown of your patterns, concrete suggestions for improvement, and a live AI chat to help you rewrite your weakest prompts. Sign in and your scores are tracked over time on a dashboard so you can see whether you're getting better at asking better questions.

How we built it

The core of AskBetter is a client-side TypeScript analysis engine that runs entirely in the browser. A user pastes a ChatGPT, Gemini, or Perplexity share link, and the engine parses out user messages, classifies each by intent (delegation, curiosity, collaborative, verification), scores them across six quality dimensions using a flag-based rubric, detects behavioral patterns, and generates suggestions. On top of that, we built a Groq-powered live chat feature so users can talk through their results with an AI assistant that has full context of their analysis. The backend is a small Express server that proxies Groq's streaming API and also handles AI chatbot share link fetching via HTTP Fetch or Puppeteer. We added Supabase for auth and a dashboard that tracks your scores over time, so you can see whether you're actually improving.

Challenges we ran into

The biggest challenge was calibrating our rubric and analysis so it felt fair. Early scoring versions penalized short prompts too harshly and rewarded long prompts just for being long. We went through several rounds of testing with real ChatGPT conversations to tune the baselines, bonuses, and penalties until the scores matched human intuition. Additionally, the overall summary feature was difficult to implement in a way that seemed neither too harsh or too nice. We had to test with multiple types of scoring configurations to ultimately get one that provided useful feedback to the user.

Accomplishments that we're proud of

We shipped a full-stack app with auth, a dashboard, and a live AI chat in under 24 hours. We are most proud of our usage of new Kiro features we had no experience using prior. Both agent hooks and steering files were new territory for us, but these ended up revolutionizing our product development. Without these, much more of our development would have had to be manual, and would have taken a lot longer.

What we learned

Building a rule-based analysis engine that feels meaningful without any LLM behind it was harder than expected. We had difficulties getting a nicely balanced scoring rubric, with summaries that weren’t too harsh but also not too nice. Learning how to implement that into our scoring metrics provided strong learning experiences for all of us. We also learned how much UI and UX matters for a tool like this. We spent a lot of time making sure our pages were clean and cohesive, and also determined that the data was displayed in an informative way. From this experience, we learned that there is much more to a project than just code functionality.

What's next for AskBetter

There are several features we would have liked to implement in AskBetter if we had more time. First, we would like to add a section where you can directly paste in a transcript between a user and a chatbot, so a link is not necessary. Second, we would like to expand our product to use more AI platforms than the three (ChatGPT, Gemini, and Perplexity) that we currently have implemented. Finally, we would like to implement a feature that takes advantage of ChatGPT’s “Export Data” feature, which exports all conversations in a user’s chat history, which could be used to allow our product to analyze many chats at once.

Built With

• eslint
• express.js
• fly.io-(deployment)
• groq-api-(llama-3.1-8b)
• javascript
• kiro
• lucide-react
• node.js
• postgresql
• prettier
• puppeteer
• react-19
• react-router-v7
• recharts-v3
• rls
• row-level-security)
• server-sent-events-(sse)
• supabase
• supabase-(auth
• tailwind-css-v4
• typescript
• vercel-(deployment)
• vite-8