description:The AI Coding Group is part of the AI Engineering organization focused on user-facing AI-powered coding features.
---
## Team Overview
## Overview
The AI Coding Group develops AI-powered features that help developers write, understand, and improve code. We focus on creating intelligent tools that enhance the developer experience across GitLab.
The AI Coding Group develops AI-powered features that help developers write, understand, and improve code. The group is organized into functional teams, each focused on a specific area.
## Our Projects
## Functional Teams
The AI Coding team owns and is actively working on the following projects:
1.[Code Review](code-review/) - Code review organization spanning the Code Review Workflow and Duo Code Review features.
1.[DAP Events](dap-events/) - Event platform and trigger system that enables Duo Flows to run automatically.
1.[DAP Repository Flows](dap-repository-flows/) - DAP onboarding and background agents that keep repositories healthy and up to date.
-[Code Suggestions](/handbook/engineering/ai/ai-coding/code_suggestions/) - AI-generated code completion and generation within IDEs.
-[Duo Code Review](/handbook/engineering/ai/ai-coding/duo_code_review/) - AI-powered code review assistance and insights.
-[Duo Context Exclusion](/handbook/engineering/ai/ai-coding/duo_context_exclusion/) - Filtering of sensitive or irrelevant code context.
-[Codebase Semantic Indexing](/handbook/engineering/ai/ai-coding/codebase_semantic_indexing/) - Advanced code search and discovery capabilities using embeddings.
-[GitLab Events Platform](/handbook/engineering/ai/ai-coding/event_platform/) - Enable GitLab Duo Flows to execute automatically.
-[Code-related Slash Commands](/handbook/engineering/ai/ai-coding/slash_commands/) - Interactive Duo Chat commands including `/explain`, `/refactor`, `/tests`, and `/fix`.
-[Repository X-Ray](/handbook/engineering/ai/ai-coding/repository_xray/) - Repository analysis and metadata extraction for enhanced code suggestions context.
-[AI Assisted Service](/handbook/engineering/ai/ai-coding/ai_assisted_service/) - Core AI infrastructure and services supporting our features.
## Feature Stewardship
### Evaluation and Testing
AI Coding is responsible for evaluations across all our features, which includes:
- Creating datasets in LangSmith and registering them in the [Datasets repository](https://gitlab.com/gitlab-org/modelops/ai-model-validation-and-research/ai-evaluation/datasets/-/blob/main/doc/guidelines/register_dataset.md#registration-process). We also have some [Code Creation Datasets](https://gitlab.com/gitlab-org/code-creation).
- Creating evaluators in [Centralized Evaluation Framework](https://gitlab.com/gitlab-org/modelops/ai-model-validation-and-research/ai-evaluation/prompt-library)
- Running evaluations
## Contact Us
Use this information to connect with the AI Coding group:
| Support | [TBD](/handbook/support/support-stable-counterparts/) |
-[Feature Stewardship](feature-stewardship/) - Features under AI Coding stewardship while permanent ownership is determined, including Code Suggestions, AI Assisted Service, and Duo Context Exclusion.
## How We Work
For information on how the team works including onboarding, time off, issue boards, meetings, and more, please refer to the [How We Work](/handbook/engineering/ai/ai-coding/how-we-work/) page.
For information on how the team works including onboarding, time off, issue boards, meetings, and more, please refer to the [How We Work](how-we-work/) page.
The Code Review organization within AI Coding is focused on making code review faster and higher quality through AI. The organization spans both the core code review workflow and AI-powered Code Review features, and is working toward a unified team structure.
The organization currently has two functional teams:
1.[Code Review Workflow](workflow/) - Core code review workflow features, moving from the Create stage into AI Coding.
1.[DAP Code Review](dap/) - AI-powered DAP Code Review features for merge requests.
Duo Code Review provides AI-powered assistance during the code review process, helping developers understand changes, identify potential issues, and improve code quality.
The DAP Code Review team focuses on [Duo Code Review](https://docs.gitlab.com/user/project/merge_requests/duo_in_merge_requests/), the AI-powered code review feature that provides automated insights and feedback on merge requests. This team is part of the broader Code Review organization within AI Coding.
## Features
@@ -14,6 +14,15 @@ Duo Code Review provides AI-powered assistance during the code review process, h
- Custom instructions for tailored review feedback
- Integration with GitLab's merge request workflow
-[Duo Code Review Monitoring Dashboard](https://log.gprd.gitlab.net/app/r/s/xVFdB) - latency for summarize review, summary merge request, and Duo Code Review (DCR); error rates, metrics for DCR comments, DCR custom instructions, and more (Kibana)
@@ -23,6 +32,6 @@ Duo Code Review provides AI-powered assistance during the code review process, h
The Code Review Workflow team is part of the AI Coding organization, having moved from the Create:Code Review group. The team focuses on AI-powered features that help developers get faster, higher-quality code reviews.
Codebase Semantic Indexing enhances code search and discovery by understanding the meaning and context of code, rather than relying solely on keyword matching. This enables developers to find relevant code more effectively and discover related functionality across large codebases. This is done by indexing projects and creating code embeddings which are searchable.
- Build semantic understanding of code content and purpose
- Enable natural language queries for code search
- Improve code discovery and reusability
- Support intelligent code recommendations
- Make the search available through the Duo Agent Platform
## What are Embeddings?
Embeddings are numerical representations that capture the semantic meaning of text or code. When we process a code repository, we break it into smaller, manageable chunks (such as functions, classes, or code blocks) and convert each chunk into an embedding — essentially a vector of numbers that represents its meaning and context. These embeddings allow us to understand not just exact keyword matches, but semantic similarity between code segments. For example, a search for "user authentication" could surface relevant results even if the actual code uses terms like "login validation" or "credential verification." By storing these embeddings in a searchable format, we enable developers to find relevant code based on intent and functionality rather than relying solely on exact text matching, making code discovery more intuitive and comprehensive.
