How AI Helps Write Unit Tests: A Wildberries Case Study
Wildberries has shared a practical case study of deploying a local AI agent in Android development. On the first attempt, the model wrote tests that did not compile and choked on context. After RAG setup, Gradle log filtering and a system of three OpenCode subagents, it began producing working unit tests. On local hardware this took about 19 minutes, and on more powerful machines 5–10 minutes.
AI-processed from Habr AI; edited by Hamidun News
How to use AI for unit tests: Wildberries shared its experience of how a local LLM, over two months, learned to write compilable tests instead of hallucinations and even to fix code review comments.
How long did it take
Wildberries' local AI to write working tests? About two months: at first the neural network made mistakes and hallucinated, but after reconfiguration it started producing compilable tests and even fixing review comments on its own.
Can a neural network be used to write code tests?
Yes: in Wildberries' experience, a local LLM learned within two months to write compilable tests and fix review comments.
How do you introduce AI into development?
According to Wildberries' practice: at first the neural network made mistakes and hallucinated, but after reconfiguration it started producing working code.
Can a neural network write unit tests?
Yes: in Wildberries' experience, a local LLM learned to write compilable unit tests for Android and even to fix review comments.
How long does it take to train a neural network to write tests?
According to Wildberries' experience, the local LLM needed about two months to get past errors and hallucinations and start producing compilable tests.
Key Questions Answered
How long did it take Wildberries' local AI to write working tests?
About two months: at first the neural network made mistakes and hallucinated, but after reconfiguration it started producing compilable tests and even fixing review comments on its own.
Can a neural network be used to write code tests?
Yes: in Wildberries' experience, a local LLM learned within two months to write compilable tests and fix review comments.
How do you introduce AI into development?
According to Wildberries' practice: at first the neural network made mistakes and hallucinated, but after reconfiguration it started producing working code.
Can a neural network write unit tests?
Yes: in Wildberries' experience, a local LLM learned to write compilable unit tests for Android and even to fix review comments.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.