You know that moment in a job interview when someone asks you a question and your brain just... empties? You have prepared for hours. You know the answer. But under pressure, the words vanish.
I built TalkBuoy to fix that.
What It Actually Does
TalkBuoy is a PWA that sits on your phone or tablet during an interview, presentation, or any high-pressure conversation. It listens to what the other person says, processes it through AI, and silently displays suggested talking points on your screen.
You glance down, see a structured response with key points you should hit, and weave them into your own words. The other person never knows.
Think of it as a teleprompter, but one that writes itself in real time based on what is actually being discussed.
How It Works Under the Hood
The app captures audio through the device microphone and runs it through speech recognition to convert what the interviewer says into text. That text gets sent to an AI model that understands the context of the conversation and generates relevant talking points.
The key technical challenges were:
Voice filtering. You do not want the app responding to your own voice. TalkBuoy distinguishes between the user and the other speaker, filtering out your speech so suggestions only respond to the interviewer's questions.
Speed. In a real conversation, you have maybe 2 to 3 seconds of natural pause before silence gets awkward. The entire pipeline from audio capture to displayed suggestion needs to complete in that window. Every millisecond matters.
Smart auto-mute. While you are reading the suggestions, the microphone needs to behave. TalkBuoy silences the mic during reading so the other person never hears anything unusual from your end.
Context awareness. The AI does not just answer the current question in isolation. It maintains context from the entire conversation so suggestions build on what has already been discussed. If you mentioned a project earlier, and the interviewer follows up on it, the AI references your earlier points.
The Use Cases That Surprised Me
I originally built this for job interviews. That is still the primary use case, and it handles behavioral questions, technical questions, and panel interviews well.
But users started using it for things I did not expect:
Public speaking Q&A. Speakers use it during the question period after presentations. You can nail a prepared talk but freeze when someone asks something unexpected. TalkBuoy fills that gap.
Sales calls. Sales reps use it to handle objections in real time. When a prospect raises a concern, suggested responses appear instantly with relevant data points and rebuttals.
Language practice. People learning a second language use it as a conversation partner. It helps them formulate responses in the target language when they get stuck during actual conversations.
Customer service. Support reps dealing with complex product questions get suggested answers without putting the customer on hold to search a knowledge base.
Podcast interviews. Hosts use it when interviewing guests on unfamiliar topics. It suggests follow-up questions based on what the guest just said, leading to better, more natural conversations.
The Ethics Question
Every time I show this to someone, the first reaction is either "that is genius" or "that is cheating." Fair enough, let me address it.
TalkBuoy does not put words in your mouth. It suggests talking points. You still need to understand the material, speak naturally, and think on your feet. If you have zero knowledge of a topic, a list of bullet points in your ear will not save you. The interviewer will hear the difference.
What it does is prevent the specific failure mode where you know the answer but cannot access it under pressure. That is a performance anxiety problem, not a knowledge problem. Most people who use it say they stop needing it after a few interviews because the practice with the safety net builds genuine confidence.
It is the same principle as training wheels. You use them until you do not need them.
Privacy
This was non-negotiable from day one. No audio is stored. No conversations are saved to servers. Everything is processed and discarded. There is no recording, no transcript history, no data mining.
The app works as a PWA, which means it runs in the browser with no app store installation required. It works on phones, tablets, and laptops. Place your device on the table or prop it up where you can glance at it naturally.
What I Learned Building It
Latency is everything in real-time AI. A suggestion that arrives 5 seconds late is useless. I spent more time optimizing the pipeline speed than on any other feature.
People do not read paragraphs under pressure. Early versions generated full paragraph responses. Nobody could process that while maintaining eye contact and a conversation. Bullet points with 5 to 8 words each turned out to be the right format.
The voice filtering problem is harder than it sounds. Distinguishing between two voices in the same room, especially through a phone microphone with room echo, required a lot of iteration. Getting this wrong means the app starts responding to your own answers, which creates a bizarre feedback loop.
PWA was the right choice. No app store approval process, no platform restrictions, instant updates. Users just visit the URL and it works. For a tool that people might need on short notice before an interview, eliminating friction was critical.
Try It
TalkBuoy is live and free to try. If you have an interview coming up, a presentation to give, or you just want to see how real-time AI coaching feels, give it a shot.
Would love to hear what you think, and if you have ideas for other use cases I have not considered.
What is the worst interview freeze moment you have experienced? Drop it in the comments.
Top comments (0)