You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
FlutterASTest is an automated tool that looks at your code and writes tests for Flutter applications. When mobile apps get big writing widget tests takes up a lot of time for developers. I built FlutterASTest to fix this problem. FlutterASTest works by combining code analysis with Large Language Models. FlutterASTest takes raw Dart code. FlutterASTest turns that code into an Abstract Syntax Tree (AST). Then FlutterASTest maps out how things like widget trees, state variables, callbacks and navigation flows all connect. Finally FlutterASTest puts all this info into a Program Knowledge Graph (PKG).
Project Goals:
Automate Widget Test Generation: I want to close the gap, between program analysis and AI-powered code generation. This will allow developers to convert Flutter codebases into widget tests that use the syntax automatically.
Enhance Semantic Context Accuracy: I plan to use Abstract Syntax Trees (AST) and Program Knowledge Graphs (PKG) to understand relationships. By looking at widget hierarchies, state transitions, callbacks and navigation flows I can prevent LLM hallucinations that happen when code inputs are unstructured.
Establish Rigorous Empirical Evaluation: I will create an automated execution and coverage pipeline. This pipeline will track how often the code compiles how many tests pass the line coverage and how efficient the executions are.
Open-Source and Disseminate: I intend to improve the framework’s security sandboxing and dependency management. Then I will release the project under an open-source license. Share the research findings.
How They Will Be Achieved:
Systematic Pipeline Execution: I will expand the analysis and PKG extraction modules. This helps turn screen architectures into token-optimized context blocks that're easy for an LLM to use.
Automated Test Execution Engine: I am adding a test-running loop that uses flutter test inside an environment. This loop will check the generated files catch any compilation or assertion errors and make reports.
Comparative Benchmarking: I will test the framework, against test baselines and raw LLM workflows. This will help me show that the AST + PKG approach is better reduces overhead and makes widget tests more effective.
We will use the grant funding to focus on three tasks needed to grow, secure and release the FlutterASTest framework.
API Token Infrastructure & LLM Inference Costs: To make the evaluation pipeline bigger we need to process thousands of AST tokens, state graphs and test files. The grant funding will pay for the LLM API inference costs that we need to run benchmarking and compare the results, against raw baseline workflows.
Sandbox Security & Execution Environments: We need to build and strengthen the execution container layer, which's the flutter test sandboxing. This flutter test sandboxing will let us safely run and check AI-generated Dart code without any risk, to the host machines or CI/CD runners.
Open-Source. Documentation: We want to move the repository from a research project to an open-source tool that anyone can use. This means we will use the grant funding to set up secure CI/CD pipelines run automated vulnerability scanning and write documentation and create datasets for the release.
Team Structure:
FlutterASTest is a one‑person open‑source project that I manage alone. I am the person who designs, builds and studies FlutterASTest from the idea all the way through technical details and final testing.
Track Record & Related Experience:
I am a software engineer and developer with a Bachelor of Science in Software Engineering from The Islamia University of Bahawalpur. Because of my background I have a lot of experience writing code checking code and building structures that work on different platforms using Flutter, Dart and backend setups.
My work on FlutterASTest comes from my engineering experience in analyzing code working with Abstract Syntax Tree (AST) parsing and making models of programs. I built the base system of FlutterASTest by myself. This includes the AST, widget and state analysis parts, the creation of the Program Knowledge Graph (PKG) which shows hundreds of nodes and connections, on each screen and the prompt creation that uses tokens well. I have shown that I can finish projects when I start from nothing.
Likely Causes of Failure:
Context Parsing Edge Cases: Flutter applications often use third-party state-management libraries like Provider, Bloc or Riverpod. These apps also build custom widget structures. If the static analyzer does not follow dependency-injection paths the Program Knowledge Graph (PKG) might lose important context.
LLM Hallucinations in Mock Generation: Large Language Models sometimes make up wrong method signatures or wrong mock dependencies when the Large Language Models create test code, for third-party packages. This can lead to compilation errors when the tests run.
Likely Outcomes & Mitigation:
Outcome if Project Fails to Achieve Full Automation: If automated test compilation rates drop low the framework can only work as a half-automated helper. It will not be a generator. If this happens a developer will still need to step in to connect the complicated mocks.
Mitigation Strategy: To fix this the project uses token tuning and a clear feedback loop. This helps us look at errors step by step. We can then improve parser rules. Get rid of mock or syntax failure patterns.
$0 (Zero USD).
Over the 12 months I have built and paid for FlutterASTest all by myself. I have not taken any money from investors, grants or big companies. I have also not used any funding to keep FlutterASTest running. I pay for all the development, the servers and the computing costs using my personal money. I am a researcher. I am the only person maintaining FlutterASTest.
There are no bids on this project.