SpeakClear

• 

  Targeted Practice: Generating an official-style IELTS Part 2 cue card.

• 

  SpeakClear Dashboard: Start your simulated IELTS speaking test.

• 

  Instant Evaluation: Receiving an accurate, AI-generated overall Band Score.

• 

  Deep Dive: Detailed feedback on Grammatical Range and Lexical Resource.

• 

  Simulated Test Environment: Recording a response hands-free.

• 

  Performance Tracking: Monitoring score improvements over multiple sessions.

• 

  Performance Tracking: Monitoring score improvements over multiple sessions.

• 

  Actionable Metrics: Analysis of Fluency and Coherence against examiner rubrics.

• 

  Line-by-Line Corrections: Identifying exact mistakes and offering natural alternatives.

Inspiration

When I was preparing for my own IELTS exam, I faced a major roadblock: finding qualified speaking partners. While I could occasionally find someone to practice with, they could not give me an accurate band score or the specific, actionable feedback I needed to actually improve. I realized that without realistic evaluation, test takers are just guessing at their progress. That frustration sparked the idea for SpeakClear. I wanted to build the accessible, highly accurate speaking coach I wish I had during my own preparation.

What it does

SpeakClear is an AI powered study assistant designed specifically for IELTS candidates. It simulates the speaking test environment, focusing heavily on targeted cue card practice, and provides instant, accurate band scoring along with detailed feedback on vocabulary, grammar, and fluency.

How I built it

The app is built with Next.js and TypeScript, styled with Tailwind CSS. Speech-to-text runs entirely in the browser using the free Web Speech API, so there's no transcription cost or API key needed for that part. The scoring itself is powered by Groq's free API, which serves fast inference on open-source models — I used this instead of a paid model to keep the whole project free to run and build. It's deployed on Vercel, connected directly to GitHub.

Challenges I ran into

The biggest one: my server kept crashing with a 502 error whenever a student finished speaking. I assumed it was an API key problem at first, but after checking Chrome DevTools' Network tab and my terminal logs, I found the real cause — the model name I'd chosen didn't actually exist on Groq anymore, so every request was failing before it even got scored. Once I found the correct current model name, feedback started working end-to-end.

Accomplishments that I am proud of

I am incredibly proud of taking a personal frustration and turning it into a fully functional MVP (Minimum Viable Product) that can genuinely help other students achieve their target band scores without needing an expensive private tutor.

What I learned

How to debug a Next.js API route by actually reading error logs instead of guessing, how free and paid AI APIs differ in practice, and how to structure a feedback loop that takes messy real speech and turns it into something a student can act on.

What is next for SpeakClear

The next steps include expanding the platform to cover all three parts of the IELTS speaking test, adding real time audio processing for a fully hands free conversational experience, and eventually expanding the diagnostics to cover IELTS Writing Tasks.

Built With

• api
• css
• express.js
• gemini
• html
• javascript
• node.js
• web-speech-api