SmartTalk

• 

  Questions

• 

  Starting Page

• 

  Gemini Feedback

Inspiration

We have all been there. The racing heart, blank minds, and the sudden inability to explain how two-pointers work. Tech interviews are very stressful and anxiety-inducing. I built SmartTalk to turn that anxiety into success. I wanted to move away from the basic Leetcode-style problem-solution approach and create a more robust feedback system using Gemini-3.

What it does

SmartTalk is an AI-powered simulator for tech interviews.

• Infinite Scenarios: Using Gemini-3, it generates unique problem sets and tests your ability to write algorithms and good code. Users can generate an infinite number of coding quizzes to test their abilities.
• Granular Feedback: After solving each question, SmartTalk provides feedback based on your code's efficiency, style, and what an interviewer might take away from your submission.

How we built it

• Brain: I integrated Gemini-3 Pro for its deep reasoning abilities, enabling users to get insightful feedback on their coding submissions.
• Stack: I used FastAPI, JavaScript, and React.js to create the dynamic frontend and a quick backend.
• Memory: I used a problem pool to speed up problem generation.

Challenges we ran into

• The biggest challenge I ran into was formatting. Gemini would sometimes return the wrong format, resulting in the code failing to render on the website.
• Gemini also ran slowly sometimes, resulting in long wait times when getting feedback.

Accomplishments that we're proud of

• Quick feedback that helps users gain deeper insights
• Creating the whole system on my own

What we learned

• I learned that LLMs can be quite faulty even with strict instructions. It showed me that I might need to use fine-tuning to get better results.

What's next for SmartTalk

• I want to add timer features as well as the ability to interact with a "digital interviewer" so users can also be emotionally ready for interviews.

Built With

• fastapi
• gemini
• javascript
• python
• react