Building AI speaking practice: notes from LangExt
Why we are building LangExt, an app that turns your own material into personalised speaking practice for exams and interviews, and what we are learning.
Speaking is often the hardest part of preparing for a language exam or a job interview. You can read and write alone, but speaking usually needs a partner, a teacher or a mock interviewer, and that is not always available or affordable. LangExt, a product we are currently developing, aims to make realistic speaking practice available whenever you need it.
The problem we want to solve
People preparing for a speaking exam or an interview tend to practise with generic question lists. Those questions rarely match their real situation: their field, their experience, the job they are applying for, or the topics their course covers. The result is practice that feels useful but does not prepare them for the questions they will actually face.
The second problem is feedback. Practising alone, you do not know whether your answer was clear, complete and appropriate. Without feedback, it is easy to repeat the same mistakes.
The idea behind LangExt
LangExt starts from the user's own material. A learner can upload documents such as a CV, a job description, course notes or a list of topics. The app then generates questions that are personalised to that material, together with suggested answers the learner can study and adapt.
When the learner is ready, they take a mock test or a mock interview. They answer out loud, just as they would in the real situation. At the end of every session, LangExt evaluates each spoken answer, so the learner knows what worked and what to improve.
Design principles
While designing LangExt we set a few principles:
- Personal over generic. Questions should reflect the learner's own material and goals.
- Speaking first. The core activity is answering out loud, not reading or typing.
- Feedback on every answer. Each answer gets its own evaluation, not just a single overall score.
- Respect for user material. Uploaded documents belong to the user and are used only to generate their practice.
What makes it technically interesting
LangExt combines several pieces. Documents need to be processed so that the app understands their content. Language models generate questions and suggested answers, which must be relevant and appropriate for the exam or interview type. Speech has to be recorded reliably on mobile devices and turned into text. Finally, the evaluation must be useful, specific and consistent.
Each step has its own challenges. Generated questions must not drift away from the source material. Speech recognition has to cope with different accents and with learners who are still building confidence. Evaluations must be helpful without sounding harsh, and specific enough that the learner knows what to change.
We treat AI models as components inside a carefully designed product, not as the product itself. Clear prompts, checks on the output and a good user interface matter as much as the model behind them.
Who it is for
We are designing LangExt for learners preparing for speaking exams, job seekers preparing for interviews, and teachers or training centres who want to give their students more practice time. The common need is the same: realistic practice with feedback, at a time that suits the learner.
Where we are
LangExt is in development. We are building the core flow: upload material, generate personalised questions and suggested answers, run a mock test or mock interview, and evaluate each spoken answer. We will share more as the product matures.
What this means for your business
The same building blocks behind LangExt, such as document processing, language models, speech on mobile and structured feedback, can support many other products: training tools, customer service quality checks, or internal knowledge assistants. If you are considering an AI feature for your product, or are interested in LangExt itself, talk to us. We are happy to share what we have learned.
Related posts
November 12, 2026
6 min read
Building software with AI: how we work at vnLOOP
AI drafts code, tests and reviews; people set the goals, approve the work and stay accountable. This is how we bring AI into everyday development without trading away quality.
November 12, 2026
6 min read
In, on or out of the loop: deciding what to hand to AI
Human in the loop, on the loop or out of the loop? Four questions for choosing how much to hand to AI, task by task.
November 4, 2026
4 min read
Automation for small teams: where to start
Small teams lose hours to repetitive work. Here is how we identify what to automate first, from bots and scrapers to data pipelines and AI features.