Test Automation

AI-Enabled Regression Test Optimization

Intelligently reducing regression-test execution effort while preserving defined coverage and confidence.

Introduction

  • Hearing-aid firmware applies more gain to quiet sound than to loud, compressing everyday sound into the range a wearer can still use. A change anywhere in that firmware can degrade the behaviour in ways inaudible on a bench but obvious to a wearer.
  • A development board runs the firmware; a host streams audio to it over Bluetooth Low Energy, captures what comes back, and measures the result. The board serves one connection at a time and each test occupies it exclusively, so running the whole suite on every build is expensive.
  • The framework reads the state of the board, scores how likely the change under test is to have broken something, and generates only the tests the evidence supports — raising the intensity and re-running on any failure, recovering the coverage the first pass deliberately omitted.

Engineering Demonstrated

  • Risk-based test selection (weighted threshold model)
  • AI planning layer with schema-constrained LLM and offline fallback
  • Dynamic pytest generation (AST-validated)
  • Failure-driven escalation and re-run under stress
  • BLE / GATT transport to nRF52840 / nRF5340 (Zephyr, Cortex-M4F)
  • On-device WDRC audio DSP verification (SNR loopback)
  • JUnit XML / CI-compatible reporting

Where This Can Help

  • Embedded firmware regression testing
  • Hearing-aid and audio DSP validation
  • Reducing regression execution time and effort
  • BLE device test automation
  • Risk-prioritised test suites in CI pipelines
  • Coverage-preserving test reduction

Open Source

Explore the source code and documentation for the AI-Enabled Regression Test Optimization project.

View Source Code

Have a similar engineering challenge?

Talk to Prajnaana about your problem.

Contact Prajnaana