AI Agent · Consumer Media
Glimpse
An interruptible AI media companion that turns videos and conversations into talk-back experiences.
Overview
Glimpse is an AI media platform that turns videos and guided conversations into interactive, talk-back experiences. Users can converse with AI-generated media in real time and even interrupt it mid-response, like a natural conversation, while a warm, emotionally intelligent companion guides them through preserving and revisiting meaningful memories. It combines video processing, voice cloning, and live speech dialogue into one subscription product.
The challenge
Traditional AI media is one-way and rigid: you watch a video or wait for a full AI response with no ability to steer it. There was also no gentle, structured way to capture and re-experience memories of loved ones that would otherwise be scattered or lost. The product needed to make media conversational, interruptible, and emotionally safe.
What we built
- A three-tier system: a Nuxt 3 client, a legacy Vue frontend, and a Node/Express API on MongoDB, deployed on Firebase Hosting with a custom API domain.
- Real-time, interruptible AI dialogue over WebSockets, streaming Google Cloud Speech-to-Text with barge-in handling so users can talk over and interrupt the AI naturally.
- A full AI media pipeline covering video ingestion, audio extraction via FFmpeg, transcription with Whisper and Google Speech, caption scraping, and GPT-driven video analysis.
- A voice cloning service that synthesizes speech in a target voice, plus an emotionally intelligent, guided-conversation engine driven by OpenAI with structured prompts.
- A productized SaaS layer with Stripe subscriptions, coupons, quota metering, Google OAuth and JWT auth, transactional email, Sentry monitoring, and S3 storage.
Results
Glimpse delivers a genuinely conversational, interrupt-capable AI media experience that feels closer to talking with a person than to using a chatbot. Its warm, guided flow lowers the barrier to preserving memories compared with a blank-page interface. It shipped as a complete commercial product, subscriptions, quotas, and billing included, making it monetizable from launch rather than a proof of concept.
