Gemini 3 Hackathon · Chrome Extension

Confused by a subtitle?
Just ask.

DocentAI is your AI docent for Netflix — like a museum guide, but for K-drama storylines. Hit Ctrl+E on any subtitle and get the full context instantly.

📦 Install Extension 🎬 Watch Demo
🏪 Chrome Web Store version coming soon.
The Problem

Watching Netflix shouldn't
feel like homework.

Complex plots, timeline jumps, and cultural references — even native speakers get lost.

😩 Before
🤔

Who is that again?

Big casts and flashback scenes make it impossible to keep up.

⏸️

Pause → Google → Resume

Searching mid-scene totally kills the vibe.

😵

Lost the plot

Forgot what happened 3 episodes ago? Same.

✨ With DocentAI
🎯

Instant context

AI tells you exactly who's who, referencing past scenes.

⚡

Ctrl+E — done.

One shortcut, no tab-switching. Under 2 seconds.

🧠

Remembers everything

Links relevant past episodes and scenes automatically.

How it works

Three steps. That's it.

No complicated setup. Works right inside Netflix.

1

Watch Netflix

Play anything. DocentAI runs quietly in the background.

2

Hit Ctrl+E

See a confusing subtitle? Press the shortcut or tap 💡.

3

Get the answer

Full context — characters, timeline, references. Done.

Demo

See it in action

Architecture

Two repos, one seamless experience.

Frontend handles Netflix. Backend runs the AI. Clean split = fast, secure, scalable.

DocentAI System Architecture
🎨

docentai-ui

Chrome Extension (Manifest V3). Grabs subtitle context, renders the UI, and calls the backend over HTTPS.

View repo →
🤖

docentai-core

FastAPI on GCP Cloud Run. Runs Gemini 3, collects references via Search Grounding, and caches results for speed.

View repo →
Why split? Separating UI from AI keeps things fast, secure, and independently scalable.
Installation

Up and running in 2 min.

Manual install for now — Web Store version is on its way.

✅

Backend is already live on GCP — no server setup needed. Just install the extension. (Check API status)

📦 Option 1: Download Pre-built (Recommended)

Easiest way to get started — no build required.

1

Download the release

Go to GitHub Releases and download docentai-ui-dev-v1.0.0.zip

2

Extract the zip file

Unzip the downloaded file to a folder on your computer.

3

Load in Chrome

Go to chrome://extensions/ → Enable Developer mode → Click Load unpacked → Select the extracted folder

4

Start watching!

Open Netflix, play anything, and press Ctrl+E (⌘+E on Mac). That's it 🎉

🛠️ Option 2: Build from Source

For developers who want to customize or contribute.

1

Clone the repo

# Terminal git clone https://github.com/tnfhrnsss/docentai cd docentai
2

Build (dev mode)

python build.py --mode dev
3

Load in Chrome

Go to chrome://extensions/ → Enable Developer mode → Click Load unpacked → Select build/extension/

4

Start watching!

Open Netflix, play anything, and press Ctrl+E (⌘+E on Mac). That's it 🎉

Features

More than just a translator.

🎯

Context-Aware

Understands the full story flow, not just the current line.

📸

Screen Capture

Optionally capture the screen so AI can read the scene visually too.

🔊

Non-Verbal Cues

Sound effects and action descriptions are sent along with subtitles.

⚡

Smart Caching

Repeated questions? Cached results come back in 25ms.

🌐

Multilingual

Korean & English. UI switches based on browser language.

🔒

Privacy First

No personal info collected. Only video metadata and subtitles.

Tech Stack

What's under the hood

Google Gemini 3 Flash Search Grounding Function Calling Chrome Extension (MV3) FastAPI Multimodal Analysis Smart Caching GCP Cloud Run SQLite JWT Auth i18n
FAQ

Got questions?

Yes — completely free right now. It's in active development during the hackathon. Premium tiers based on API usage may come later.
Netflix only for now. YouTube, Disney+, Wavve, and more are planned.
Absolutely. No names, emails, or browsing history are collected. Only video metadata (title, episode number) and subtitle text are sent — everything encrypted in transit.
Gemini 3 + Search Grounding pulls in real reference data, so accuracy is high. Occasional misses are possible — we keep improving based on feedback.
Internet is needed for new queries, but previously cached explanations are available offline.
Clean separation of UI and AI. The backend uses a smart 2-step process: Search Grounding runs once when you start a video (cached), then each explanation reuses those references — keeping it fast and cost-efficient.

Ready to try it? 🎬

Install in 2 minutes. No sign-up needed.

📦 Installation Guide GitHub →