MockTalk: Real-time AI-powered presentation practice app without real-time models
Published:
TL;DR
MockTalk is a real-time AI presentation practice tool for academics: think a mock PhD advisor who’s always available, listens to you present over your slides, interrupts with questions, and gives feedback. It uses multimodal turn-taking (screen capture + audio transcription) sent to LLMs on a regular loop, so it feels live without needing expensive real-time APIs. Users bring their own API keys, pick a model, and just start practicing. I built it as a side project for fun in my free time collaborating with Claude Code. This blog covers roughly why I built it, how it works under the hood, and what I actually think about AI-assisted development after living in it for a while.
MockTalk LIVE here: https://mocktalk-k3yg.onrender.com (NOTE: I hosted this on Render’s free-tier, so it does take 30-50 seconds to spin up the server.)
Fully open-sourced Github repo here: github.com/shaox192/mocktalk
This is a working progress :construction: … Check back later!