Demonstration
See it running.
The problem
Why I built it.
My students already knew vocabulary and how to build a sentence. What they could not do was hold a conversation. Most practice apps hand you disconnected phrases with no situation around them, so the moment a real person answers something unexpected, the practice does not transfer.
Intended users. English learners who have finished the grammar stage and need the next leap: speaking with a partner who reacts to what they actually said.
Design
Decisions I made and why.
01
Context over phrase drills
A phrase you repeat in isolation is a phrase you can only use in isolation. Every prompt sits inside a situation the learner can picture, and the answer they get depends on what they just said.
02
Feedback while the attempt is still warm
Correction a day later is a grade. Correction two seconds later is learning. Feedback arrives right after the recording, while the learner still remembers the choice they made.
03
Native pace on purpose
Slowed speech is comfortable and misleading. Learners who only practise against slow audio freeze in a real conversation, so the speaking partner talks normally.
Disclosure
What I built, what I would test, what it cannot claim.
What I built+
- A conversation loop where each reply follows from the previous turn instead of resetting
- Natural spoken English at native pace, so the listening practice matches real conversation
- Recording of the learner's own speech as the input to the next turn
- Immediate feedback tied to what the learner recorded, not to a generic model answer
- Situations that continue over several turns so context carries forward
What I would test next+
- A before and after speaking sample scored by someone who does not know which learner used the tool
- Tracking whether feedback actually changes the next attempt, turn by turn
- Comparing conversational practice against the same time spent on phrase drills
Limitations+
- I have not run a controlled study on whether this improves speaking ability. I do not claim measured learning gains.
- The evidence on this page is a recorded demonstration of the tool working, not outcome data.
- It was designed around my own students, so it has not been tested with learners outside that group.
- Instant feedback can feel like progress even when it is not producing it. That risk is untested here.