# 🤖 SpeakMate AI: An AI English Practice Companion I Built for a Friend
Learning a language is not only about knowing the rules. You need a place where you can actually practice, make mistakes, and understand why an answer is correct.
One of my friends is learning English, and I wanted to build something specifically around that problem.
For the Hacktoberfest 2026 Weekend Challenge, I built SpeakMate AI — a focused English practice companion for a real friend. ## 🎯 The Problem
My friend needed a simple way to practice English regularly.
A general chatbot can answer questions, but I wanted to build something more focused.
Instead of simply chatting with an AI, SpeakMate gives the learner actual exercises:
The goal was simple:
Don't just talk to the learner. Give them something to practice.
## 💡 What I Built
SpeakMate AI has three main practice modes.
### 📝 Grammar Practice
The learner chooses a difficulty level:
The AI generates a grammar exercise with multiple-choice answers.
After answering, the learner gets:
### 📚 Vocabulary Practice
The vocabulary mode introduces words with:
This makes vocabulary learning more interactive than simply reading a list of words.
### ✍️ Sentence Correction
The learner receives an incorrect English sentence.
They then choose which option correctly fixes the sentence.
After checking the answer, SpeakMate explains the grammar mistake.
## 🧠 The AI Behind SpeakMate
One of the most important parts of this project is that the AI is not just a remote API call.
I used:
The model runs in the browser using Transformers.js and WebGPU.
This means the project can experiment with local inference rather than depending entirely on a paid hosted AI API.
The open approach also makes the AI layer easier to experiment with: the model and inference setup can evolve as the project evolves.
## 🌱 Why Open Innovation Matters Here
For this project, using an open/open-weight model and local inference was important because I wanted the learning tool to be lightweight and inexpensive to experiment with. A learning companion shouldn't necessarily require a paid AI service every time someone wants to practice a grammar question.
Running the model locally also gives the project more control over how the AI component is used and makes it possible to experiment with different models and configurations.
That was one of the biggest things I wanted to explore through this challenge.
## 🏗️ How It Works
The basic flow is:
Choose Practice Mode
↓
Choose Difficulty
↓
Local AI Model
↓
Generate Exercise
↓
Answer Question
↓
Check Answer
↓
Explanation + Score
### Sentence-correction Practice
## 🛠️ Tech Stack
| Technology | Purpose |
|---|---|
| Next.js | Web application |
| TypeScript | Application development |
| Tailwind CSS | UI |
| Transformers.js | Running the AI model |
| WebGPU | Browser acceleration |
| Qwen2.5 | Language model |
## 🔗 Project
The complete source code for SpeakMate AI is available on GitHub:
View the SpeakMate AI source code on GitHub You can clone it and run it locally with:
npm install
npm run dev
Then open:
the link A browser with WebGPU support is recommended for the local AI model.
## ❤️ What I Learned
Building SpeakMate taught me that a useful AI project doesn't have to be a huge application.
The most important part was starting with a real person's problem and building around that problem.
I also learned more about:
## 🚀 What's Next
There are several things I would like to add in the future:
For this weekend, however, the goal was simple: Build something useful for one real person.
And that's what SpeakMate AI is.
Built for the Hacktoberfest 2026 Weekend Challenge — Build for a Friend.
#devchallenge #weekendchallenge #hf26challenge