Our goal is to make collaborative software development scale without losing any efficiency.
As teams get larger, it becomes challenging to maintain a consistent & efficient recipe for software development.
Code reviews are one of the main checkpoints and safeguards for team alignment today.
With code generation getting commoditized, we’ll need to allocate more efforts on the alignment layer, which is where we step in.
The current version of the system relies on two main pipelines that use deep learning models (namely language models, either instruction-tuned or completion-based).
Guideline curation & validation
Parsing & confirming guidelines to build a useful context for AI-powered developer applications.
Coding assistance
Producing or completing software-related content with context awareness.
From an architecture perspective now, here are all the system components: