01 / CODERABBITSummer 2026

Applied AI engineering at CodeRabbit.

Software Engineer Intern, Applied AI

Work completed during a summer internship across repository intelligence, model evaluation, deployment, and optimization.

Cross-repository context for code review.

Changes in one repository can affect packages, services, and downstream applications elsewhere. This work focused on identifying those relationships and making relevant repository context available to code-review systems.

2,000+repositories
232organizations
75%of routed model traffic
8–9%production cost reduction

Areas of responsibility.

The internship included repository-linking systems, evaluation infrastructure, model deployment, and prompt-cost analysis.

The system combines import graphs, SBOM data, architecture signals, and confidence ranking to identify related repositories.

IMPORT GRAPHSBOMARCH SIGNALSCONFIDENCELINKED REPOS

Automatic repository linking.

The linking system combines import graphs, SBOM and dependency data, architecture signals, and confidence ranking. It identifies related repositories without requiring a manually maintained map.

CodeRabbit engineering article ↗

Model and prompt efficiency.

DeepSeek V4 Flash handled 75% of routed file-summarizer traffic. Removing duplicated file-summary context from a production prompt reduced walkthrough-generation cost by 8–9% while meeting the existing evaluation criteria.

Company and product outcomes are presented at a high level; proprietary implementation details are intentionally omitted.