FoodLog Live

FoodLog Live turns every meal into a story you can capture and share. It sees the food in front of you, listens to your questions, and talks with you about it in real time.

Comment

Inspiration

FoodLog Live is the result of multiple hackathon attempts. I kept returning to the same idea: food has a story, and capturing it should feel like a conversation.

OpenAI Realtime, GPT-5.6, and Codex finally gave me the tools to build the experience I had imagined.

What it does

FoodLog Live can see, hear, and talk about the food in front of you.

Point the camera at a meal and start talking. FoodLog identifies what it sees, answers questions, and creates a structured food entry with nutrition and a photo. You can correct entries, review your day, and access the same food history from Codex through MCP.

How I built it

I built FoodLog Live with Flutter, OpenAI Realtime, GPT-5.6, Codex, and a Go backend running on Google Cloud.

OpenAI Realtime powers the live voice and camera conversation. GPT-5.6 handles deeper food understanding and structured tasks.

Codex was my engineering partner. It helped me build the OpenAI integrations, Go services, MCP server, deployment system, and review tooling. I used multiple Codex tasks working in isolated branches to build, test, review, and improve the product.

Challenges I ran into

Realtime voice and vision involve constant coordination. Listening, speaking, interruptions, camera frames, tool calls, and saved data all have to stay synchronized.

The other challenge was working at this scale as a solo developer. I needed Codex to move quickly without losing control of the repository. That led me to build stronger ownership, review, and merge systems around it.

Accomplishments that I'm proud of

I turned a long-running hackathon idea into a working multimodal product built around OpenAI.

The same food history now works across voice, web, and Codex through MCP.

I also built foodlog-ai, an AI review system that reviews Codex's work before it can be merged. Codex helped build the system that now checks its own output.

What I learned

I learned how useful Codex can be across the entire development process, not just for generating code.

I used the Codex desktop app to plan work, coordinate parallel tasks, and keep track of branches and pull requests. I used the CLI inside the repository to refactor code, run tests, fix review feedback, and deploy the application.

Using both together helped me move between planning and implementation without losing context.

What's next for FoodLog Live

Next I want to build shared meals, family food stories, better long-term insights, health-platform integrations, and more ways for people and agents to interact with their food history.

Built With

Share this project:

Updates