Personal project
SoundLens - AI-assisted acoustic investigation
An AI-assisted platform for investigating recordings, product variants, test conditions, and algorithm outputs using deterministic DSP evidence.
View repository
Personal project00 — Context
The problem
Engineers need one traceable workflow for investigating batches of recordings, selecting regions, comparing variants, and checking the evidence behind an interpretation.
01 — Role
My contribution
- 01
Designed the product direction, backend/frontend architecture, and technical-user workflow.
- 02
Built deterministic waveform, spectrum, metric, tonal, and region-scoped analysis.
- 03
Implemented an AI Copilot using structured tool calls over typed evidence, units, and limitations.
- 04
Created traceable reports, refusal behaviour, and automated evaluation for evidence-grounded AI-assisted analysis.
- 05
Designed the roadmap toward batch comparison, anomaly investigation, and scalable engineering workflows.
02 — Engineering
Technical approach
- 01
C#/.NET backend with a React and TypeScript frontend.
- 02
Separate deterministic DSP calculations from the AI explanation layer.
- 03
Ground every generated answer in typed evidence with explicit units, scope, and limitations.
- 04
Structure analysis capabilities as tools that can support increasingly complex investigation workflows.
03 — Quality
How I checked it
- 01
Automated backend, frontend, and workflow tests check deterministic calculations, tool contracts, and expected UI behaviour.
- 02
Evaluation cases check missing evidence, unsupported questions, scope and unit handling, and safe refusal behaviour.
- 03
Generated answers are checked against the exact analysis evidence they reference; real-user usefulness and production reliability remain separate validation work.
Continue the conversation
Discuss this work.
I can share additional context about the engineering decisions and my contribution where confidentiality allows.