Inspiration
I have studied English for many years and majored in English at university, but I still found it difficult to watch English-language TV shows or YouTube videos without subtitles.
That made me question what was missing from the way I had learned English. I began to feel that improving my listening and speaking required more exposure to authentic videos made by native speakers, gradually relying less on subtitles, and working with the language sentence by sentence.
I later subscribed to a website that allowed learners to practise with short video clips through dictation, sentence replay, speed control, and shadowing. The method was useful, but the videos were selected by the platform, so I could not choose content I genuinely wanted to watch. Looking up an expression required leaving the learning environment, and there was little opportunity for interaction or speaking practice.
I also tried existing YouTube learning extensions, including Language Reactor and Trancy. They offered useful features, but I still did not find an experience that felt simple, integrated, and complete enough for the way I wanted to learn. That led me to build Echo One.
Echo One is primarily designed for English learners in China who do not have regular access to an English-speaking environment.
Many Chinese learners begin studying English at a very young age, but after years of study, they may still struggle to speak naturally or confidently. This is often described as “mute English”: learners may know grammar and vocabulary, but lack exposure to authentic language and opportunities to use English in real interaction.
I want Echo One to help learners understand what real English sounds like, remember expressions they can actually use, and feel safer speaking. This sense of safety is especially important to me. Speaking a foreign language can feel emotionally risky, and AI can provide a more nonjudgmental space for learners who are sensitive or self-conscious about making mistakes.
What it does
Echo One is a Chrome extension that turns YouTube videos into sentence-by-sentence English learning experiences.
It works directly inside the YouTube player and currently provides:
- English and Chinese bilingual subtitles
- Automatically organized sentence-level subtitles
- Automatic pausing after each sentence
- Previous-sentence, next-sentence, and sentence-replay controls
- Adjustable playback speed
- Dictation practice
- Shadowing practice
- Context-aware AI questions about the current video
- Context-aware voice interaction for speaking practice
Instead of limiting learners to a fixed library of educational clips, Echo One allows them to learn from YouTube videos they already enjoy.
The core learning path is:
authentic input → focused listening → sentence-level practice → active speaking
How we built it
Echo One is built as a Chrome extension integrated into YouTube.
The extension works with available English subtitle tracks and restructures them into sentence-level learning units. It synchronizes these units with the video so that learners can pause, replay, dictate, and shadow one sentence at a time.
The current contextual AI features use DeepSeek because keeping operating costs manageable is important for an early-stage product. Google Text-to-Speech is used for voice output. In the future, I would like to explore integrating GPT Live to create a more natural and responsive speaking experience.
Almost the entire extension has been developed with Codex as my engineering collaborator. As someone without a traditional software-engineering background, I used Codex to understand unfamiliar parts of the codebase, turn product ideas into implementation plans, write and revise code, investigate bugs, and improve reliability.
Echo One already had its core learning and contextual AI features before OpenAI Build Week.
During Build Week, I used Codex powered by GPT-5.6 Sol to strengthen the product foundation behind those features. The main work included:
- Building backend infrastructure
- Creating a user database
- Preparing the system to associate users with future subscription plans and purchases
- Creating a foundation for future personalized learning features, including saved vocabulary and spaced-repetition review
- Adding tests
- Improving error handling
- Strengthening backend reliability and security
This work was less visible than adding another button or learning mode, but it was necessary for Echo One to grow from a standalone browser tool into a personalized learning product.
Challenges we ran into
One of the biggest challenges was building a payment system.
I spent a full day exploring an automated payment approach with Codex. The solution ultimately did not work for the product as intended, so I chose not to ship it.
At first, this felt like wasted time and wasted tokens. However, it also provided useful feedback. It showed me that payment is not only about displaying a checkout page. A sustainable system also needs reliable user identification, purchase records, plan access, and a clear connection between payment status and product permissions.
That failed attempt clarified why the user database and backend foundation had to come first.
Another continuing challenge is cost. Contextual text and voice interactions can become expensive as usage grows. This is why the current version uses DeepSeek and Google Text-to-Speech. I am trying to balance interaction quality, accessibility, and operating cost while the product is still being validated.
A further challenge is that Echo One depends on YouTube videos having a readable English subtitle track. It does not currently generate transcripts directly from audio.
Accomplishments that we're proud of
I am proud that Echo One has moved beyond an idea and is now a working Chrome extension available to real users.
Apart from myself, five users have returned to use the extension repeatedly, and I have received two direct comments saying that the product was useful. This is still a very small user base, but it is meaningful to me because it shows that the learning problem is shared by people other than myself.
I am also proud that I built almost the entire product with Codex despite not having a traditional software-engineering background.
During Build Week, I did not focus on adding another visible feature. Instead, I worked on the less visible infrastructure required for future accounts, subscriptions, saved vocabulary, learning records, and personalized review. Completing this foundation helped move Echo One closer to becoming a sustainable product rather than only a personal tool.
What we learned
The most important thing I learned is that building a useful AI product involves much more than adding a model to an interface.
The quality of the experience depends on:
- What context the AI receives
- When the interaction appears in the learning process
- Whether it helps the learner stay focused
- Whether the learner feels safe enough to speak
- Whether the underlying system can support real users reliably
I also learned that invisible infrastructure can be as important as visible features. A user may not directly see a database, tests, or backend error handling, but these systems determine whether the product can eventually support subscriptions, saved vocabulary, learning history, and personalized review.
Working with Codex also changed what I believed I could build. It did not replace product judgment. I was still responsible for deciding what problem to solve, who the product was for, how the learning experience should work, and which trade-offs were acceptable. But Codex allowed me to explore technical areas that would otherwise have been inaccessible to me.
I also learned that an unsuccessful implementation attempt is still useful when it reveals that an architectural assumption is wrong. The failed payment approach helped me better understand what the system would need before payments could work reliably.
What's next for Echo One
My next goal is to test Echo One with more Chinese-speaking English learners and understand which parts of the experience make them return.
I plan to continue developing:
- User accounts and subscription access
- Personal vocabulary collections
- Learning history
- Spaced-repetition review
- More natural contextual speaking practice
- A clearer path from listening and comprehension to active speaking
I would also like to explore GPT Live in the future, especially for faster and more natural voice interaction.
Ultimately, I want Echo One to help learners understand authentic English, remember natural expressions, and feel safer using the language before they enter a real English-speaking environment.
Log in or sign up for Devpost to join the conversation.