Apollo

Apollo

Apollo

Inspiration

Although diffusion models have advanced rapidly, the general public has a disconnect between AI and the creative arts, almost creating a zero-sum game rather than exploring artistic innovation with AI tools. We hope our project acts as a bridge to demonstrate this possibility and optimize the creator workflow end-to-end.

The content creation workflow can be daunting, encompassing idea generation, editing, analysis, optimization, sponsor relationships, and revenue generation. Apollo seeks to lower this barrier to entry so more people can join the content-creation age.

What it does

Apollo is an agentic video editor that expands its feature set to optimize creator workflows.

A full non-linear editor gives professional control over timeline editing, trimming, transitions, color correction, and multi-format export.
An AI editor agent lets users describe changes in plain English. The agent reads the state of the timeline, plans actions, and executes them automatically.
A storyboard mode turns a story prompt into a complete video. Gemini plans the scenes, Veo generates clips with visual continuity between them, and creators can refine individual clips in place before exporting to the timeline.
An analytics tab connects to YouTube via OAuth to surface performance data, then uses Gemini to generate actionable insights and AI-powered thumbnails.
A sponsor tab analyzes video content, matches topics against a database of over 100K advertisers, drafts outreach emails through Gmail, and generates sponsor-ready clips, all through an AI chat agent.

How we built it

Apollo is built on Electron with React for the desktop UI, a Python gRPC service as the AI orchestrator, and a Rust workspace for the performance-critical media engine.

All AI capabilities run through Google's Gemini 3 model family:

Flash for planning and perception
Pro for analytics and research
Pro Image for thumbnail generation
Veo 3.1 for all video generation

Data flows through SQLite caches, the YouTube Data and Analytics APIs, BigQuery for advertiser data, and the Gmail API for sponsor outreach, all connected via OAuth 2.0. The Rust engine handles timeline evaluation, proxy generation, and the FFmpeg-based export pipeline.

Challenges we ran into

The editor's feature-rich, complex nature made it difficult to model accurately and optimally.

Video generation optimizations, passing context, continuity, and logic between scenes, is a multi-step process with many moving parts. Getting the data we needed, such as sponsorship information, required developing unique solutions, including using BigQuery paired with Gemini for online lookups.

One of our biggest challenges was fitting our feature set into a 3-minute video.

Accomplishments that we're proud of

Seamless video continuity in the storyboard pipeline.
An agentic framework built on top of a Rust engine video editor.
Designing a framework to analyze data without significant manual effort.
Reducing the barrier to getting sponsorships by making the process more accessible.

What we learned

On a human level, my teammate and I worked intensively over the course of a month to build a feature-rich application we designed ourselves. This intensive process helped us learn the power of staying focused and working toward a goal we hope to complete. The project's timeframe and open creative prompt allowed us to let our imaginations run wild and think big.

On a technical level, the power of these frontier-generation models is truly incredible. They can reason multimodally and generate impressive clips and images. The inner workings of tools like FFmpeg were great to explore, as was designing the architecture of complex applications.

What's next for Apollo

We want to continue development because the features can be continuously improved. Moreover, as models advance, these editing features become more powerful. Continuous development will optimize our project and build out features that are still in progress. Ideally, we would be fortunate to meet with the AI fund teams for advice.

Thanks for taking the time.

Note: Our demo video was created using Apollo.