Lattice

Trying to be an AI/ML engineer without the proper hardware to work with.

ENITRE PROJECT IS MADE USING GPT-5.6SOL ON ULTRA

I am a 4th year student doing bachelors in Artificial intelligence, trying to build and ship real working applications that actually solve a problem rather than just being test notebooks or gimmicks. However, I was facing a lot of issue with training and testing models as my laptop does not have a proper dedicated gpu to handle such tasks. The work around was to make my projects locally on my laptop then with the help of codex package the model only into an ipynb file export all the training artifacts to colab, train the model there and import the trained versio n back to my system to wire it in, only to find out that the model has some compatibility issue during inference time. This isn't actually a very reliable or smooth workflow. This gave me the idea of Lattice. Lattice is an IDE that harnesses the power of google colab GPUs leveraging the colab MCP server into the IDE itself, allowing you to do everything that you can in a normal IDE and much more! You can easily write code manually or using codex that is built into it and train your models locally within the ide itself. codex can itself connect to colab gpus when needed do training runs, monitor/ interrupt/ edit training runs and make the entire project within the IDE itself never having to exist the environment regardless of your own systems capabilities! We did face a challenge in ensuring that you dont need an ipnyb file to train models locally. you can should be able to train any file format never having to worry about the architecture effecting the training run just as if you had CUDA on your own system. We hope to make the intelligence better. possibly giving it the ability to estimate the exact amount of gpu and vram you need for a specific task making it more cost efficient. Possibly even collaborating within openAI to introduce some LATTICE-ONLY features for codex such as the ability sleep during long training runs and ocassionally waking up to monitor progress and making any edits in real time to the training run if needed!

Built With

• ai
• developer-tools
• git
• google-colab
• ide
• json-lines
• local-first
• machine-learning
• mcp
• monaco-editor
• node.js
• openai-codex
• python
• react
• training
• typescript
• vite
• windows
• workspace-management
• wsl2