Inspiration
The most powerful software applications of the future will be the ones that can "see" the world with us and help us interpret it.
I love to drive. But when I drive, all of my attention is on the road, and I miss the boring as well as the interesting stuff I pass by. I wanted a co-pilot that could record, observe, and share back a summary with me when I was done with my drive.
That desire - to turn the "Mundane to Octane" - was the spark for this project.
What it does
At its core, OctaneLog records my drive when I start a drive, and asks Gemini for a summary when I end my drive.
How I built it
OctaneLog is an iOS application that uses on-device Vision AI and Gemini 3 to capture, analyze, and narrate your life on the road.
- The Director: Continuously records your drive and uses Apple's Vision framework to analyze scenery, lighting, and objects in real-time.
- The Editor (Gemini 3): Combines video footage with local Vision analysis to generate vivid narrative summaries of your drives. ** Gemini 3 Flash: Analyzes driving footage for visual storytelling. ** Gemini 3 Pro: Manages the "Season Arc" and deep reasoning tasks.
- Smart Processing: Intelligently uploads driving footage to Gemini when local Vision analysis detects significant events or scenery, ensuring high-quality narration
Challenges I ran into
It was extremely hard to validate changes in the real world! After all, I couldn't just take my car out for a drive every time a change was made to the code. I relied entirely on AI to build and write test cases - but all real world execution had to be manual.
There are some particularly vexing issues I have tried to bypass with my take on the app's design, such as avoiding user-generated data to be stored on any intermediate server - data goes straight from the phone directly to Gemini, without involving any server I or any other company (other than Google) owns.
Accomplishments that I am proud of
I do not know Swift! This app and its source code is entirely coded using AI tools, primarily Antigravity and Gemini 3 Pro (High).
While the app is not on App Store yet, it can be built and downloaded directly to an iPhone. I have been testing it out and works great!
Being curious, I could not help myself from not seeing the source code, but I made no edits at all! Anytime I noticed an issue - it was just a prompt in Antigravity for Gemini to figure it out!
What I learned
Despite having never developed an iOS app or coding in Swift, I was able to build a working app for myself. And the description from Gemini on the actual video drives was pretty accurate.
One other lesson from the coding sessions was to stay focused on a task - since the ability to branch off in different directions is very real with the new capabilities!
What's next for OctaneLog
The main goal would be for anyone to be able download and use the app from the App Store. But I have not yet been able to do that because I could not figure out a workflow where I was not the one guessing what would I owe to Gemini API or an interim server that could host the user generated data (video logs).
Once I can figure out an efficient way to do so - that would be the next milestone.
And then, I will continue pushing changes to the repository to add more functionality, such as a way for users to influence the current prompts to tailor Gemini's output to their taste. The prompts are currently hard-coded in the app and I want users to have the ability to give their drive logs their own taste or personality.
Log in or sign up for Devpost to join the conversation.