Icamefromthesameplace.

It wasn’t until 2020, after COVID, that I restarted learning English. I admitted my foundation was terrible and started from the basics—vocabulary, one word at a time.

Since I used Twitter every day and already read English posts, I thought I was ready. But when I first completed a reading mock in May 2021, I scored only 5.5—and it took me 3 hours.

Reading and listening improved relatively easily. But writing and speaking? I had no idea what to do.

Since I had a full-time job, I hired personal one-on-one tutors for Speaking and Writing. It made a difference. I got some techniques.

March 25, 2022 — I got my 6.5.

One of the most exciting days of my life. This was the target I set for myself: if I couldn’t achieve this, I wouldn’t go abroad.

I had already used ChatGPT to create IELTS Reading and Writing feedback GPTs. But speaking was still my main issue.

November 2023 — OpenAI released GPT with voice. In the shower, I had an idea: what if I could simulate an IELTS speaking exam?

I built the IELTS Speaking Simulator GPT that December and open-sourced it. I believed it was extremely helpful—and realized it would be an excellent example for AI since it has clear benchmarks for evaluation. Even hallucination is a feature, not a bug.

I used it daily, updated it, and released V3. After work, I’d run a simulation, get feedback, copy it to Notion. One issue: I rarely reviewed the feedback—I just practiced. Even on the same topic, I kept making the same mistakes. I don’t like memorizing...

I posted it on RedNote to help more people. It became one of my viral posts.

November 2024 — OpenAI released the real-time API. Impressive experience, just like the current product. I tried the playground and built a demo.

But it was expensive. A 2-minute conversation cost over $2.

2025 became the year of vibe coding. I gradually went from non-technical to building frontend projects with Cursor, and then Claude Code.

June 2025 — Before my first trip to the Bay Area, I updated the IELTS Speaking Simulator question bank again on RedNote. Surprisingly, this post went viral—crazier than before.

The most important lesson from that trip: build something that belongs to yourself.

After returning to Ottawa, I had a plan: build a prototype using OpenAI’s real-time API. AGI Venture Canada was hosting a vibe coding hackathon—just in time!

I built a quick demo using free Anthropic credits from my Silicon Valley trip. What a coincidence! Surprisingly, it worked.

After the hackathon, I spent two more weeks polishing and launched the standalone app.

But here’s the truth: I had thought about building a real product. However, my capability—and even Sonnet 4.0—weren’t enough. I was afraid to modify features. Fixed one bug, received three more.

So I gave up. I planned to wait for more powerful AI.

Then there was Mark.

In my first month in Canada, before a networking event, I met Mark—a non-native English speaker who sounds like a native. Former army officer, economist by trade, and even United Nations employee, currently retired full-time grandpa and part-time “citizen in action” (in his own words) who is very happy to share his life and language learning experiences with everyone seeking new skill levels. Mark’s brand philosophy: “No brand, just be yourself.”

After he learned I wanted to improve my English, he started sending me voice messages—teaching me real-world communication. He gave me mini-programs like “right here right now” and encouraged me to report my day in no more than 2 minutes of audio.

At first, I just recorded some. Not consistent.

It wasn’t until last summer that I took English more seriously. I started sending my daily report to Mark every single day—and received his feedback. He even suggested I call it “Reporter Joe Speaking” as my brand, to build confidence.