QuackBack

Inspiration

Every programmer has experienced this: asking for help and getting an answer that works, but not understanding why. While modern AI tools are great at producing solutions, they often skip the thinking process that leads to real learning. QuackBack works because it slows you down, forces you to articulate your thoughts, and exposes hidden assumptions.

We wanted to recreate that experience digitally, without shortcuts or copy-paste solutions. Our goal was to build a tool that encourages reasoning over guessing, learning over dependency, and clarity over speed, helping developers arrive at their own insights.

What it does

QuackBack is an AI-powered rubber duck for programmers. Instead of giving answers, it asks thoughtful questions that surface assumptions, clarify intent, and guide users toward their own insights. It helps developers debug, design, and reason more effectively.

How we built it

QuackBack is an AI-powered rubber duck for programmers. Instead of giving answers or code snippets, it asks thoughtful, targeted questions that surface assumptions, clarify intent, and encourage deeper reasoning. By guiding users through their own thought process, QuackBack helps developers debug issues, think through design decisions, and understand problems more clearly.

Conversations are saved so users can revisit their reasoning, reflect on “aha” moments, and see how their thinking evolved over time. The result is a tool that supports learning and problem-solving, not just quick fixes.

Challenges we ran into

Getting an LLM to not give answers was surprisingly hard. We had to carefully design prompts, enforce strict output schemas, and handle edge cases where the model tried to be “helpful.” We also learned how important it is to prevent accidental context leakage so each session stays focused and unbiased.

This was our first time working directly with AI APIs, which was both exciting and challenging, especially when testing behavior locally and debugging inconsistent outputs. It pushed us to better understand how LLMs actually behave in practice, not just in theory.

It was also our first time collaborating together, we formed the team just before the start of the hackathon. Working with a teammate across the world in a very different time zone required clear communication, trust, and careful coordination. While we were initially nervous about stepping on each other’s work, learning to divide responsibilities and integrate our changes became one of the most rewarding parts of the project.

Accomplishments that we're proud of

We successfully flipped the AI interaction model from delivering answers to asking thoughtful questions, creating a tool that encourages learning and reasoning rather than dependency. Within the constraints of a hackathon timeframe, we built a complete, fully working product with a polished user experience and a clear personality.

Along the way, we also challenged ourselves by working with new languages, libraries, and AI tooling, expanding our technical skills while still delivering a cohesive and reliable application.

What we learned

We learned that constraints are powerful. By intentionally limiting what the AI is allowed to do, we created a more meaningful experience that encourages thinking rather than dependency. Designing and enforcing those constraints taught us a lot about prompt engineering, schema validation, and making AI behavior reliable under real-world conditions.

We also learned the value of clear collaboration under time pressure. By dividing responsibilities and trusting each other’s strengths, we were able to work in parallel, integrate quickly, and keep the project moving forward. Clear ownership and communication proved to be just as important as our technical decisions.

Along the way, we gained hands-on experience with new technologies, including working with AI APIs for the first time, one of us building with Next.js, and integrating several unfamiliar libraries. Learning these tools while shipping a complete product made the experience both challenging and rewarding.

What's next for QuackBack

Next, we’d like to add conversation summaries, question-type insights, and reflection tools that help users see how their thinking evolved. Long-term, we see QuackBack as a personal thinking companion for developers who want to grow.

Built With

• gemini
• mongodb
• next.js
• react
• typescript