Test Generation

​

Overview

Tusk utilizes an optimal mix of LLM-based and rule-based techniques to generate tests that are both comprehensive and context-aware.

Our rule-based techniques like using the Language Server Protocol and abstract syntax tree enable Tusk to find relevant codebase context 3x faster than similar tools while being more reliable.

Unlike general-purpose AI copilots or chatbots, Tusk generates tests with targeted knowledge of your mature codebase and business logic and autonomously runs and iterates on the tests to ensure they are valid. Additionally, our agent uses secondary filters to make sure that tests are de-duplicated and of high relevance to the PR/MR.

These built-in mechanisms abstract away the overhead of manually writing good tests.

​

Generating High Quality Tests

We’ve designed Tusk’s test generation pipeline to replicate the process of a senior software engineer writing tests by hand, while taking into account the nuances of working with LLMs.

Here’s a high-level rundown of Tusk’s test generation pipeline: