NDA-safe · Apple via INSPYR Solutions · 2026Lead Moderator

Multi-Device AI Audio Capture Study

100+

Participants tracked

50+

Sessions moderated

Teams aligned

Research goals

01Capture naturalistic user audio across multiple simultaneous device configurations.
02Compare model and software performance against real user behavior.
03Evaluate AI audio recognition accuracy across diverse user populations.

Methods

01Structured prompt-based audio elicitation (60 min sessions).
02Multi-device simultaneous capture with parallel moderator setups.
03Diverse prompt domains to capture natural language variation.
04QA review and post-session data upload protocols.

Research process

Scoping & stakeholder alignment

Engaged with multiple managers across teams with different priorities and technical requirements. Translated competing needs into a unified study design that could serve engineering, design, and research simultaneously, without letting any one team's requirements compromise data integrity.

Protocol design

Authored the end-to-end study protocol from scratch: physical device setup with diagrams and specifications, in-house software recording configuration, parallel app workflows, step-by-step session instructions, troubleshooting guides, and QA/post-session upload procedures. Designed for replication across future phases and locales.

Prompt design & participant management

Wrote diverse prompt sets spanning numerous domains to elicit natural language variation across participants. Built a tracking system for 100+ participants and managed outreach, scheduling, and rapport, establishing trust to reduce performance bias during sessions.

Moderation & execution

Served as lead moderator, personally running 50+ one-hour sessions while maintaining study consistency and data integrity across a high-volume collection period. Trained and oversaw a second moderator running a parallel setup, doubling throughput without sacrificing protocol fidelity.

Handoff & impact

Data delivered to engineering teams to inform AI model decisions. Protocol built to serve as the foundational template for future study phases and potential expansion to international markets and additional languages.

Key research decisions

Chose structured prompt elicitation over free-form interaction to control for variability while still capturing natural speech patterns, critical for meaningful AI model comparison.

Built participant rapport intentionally before and after sessions to reduce performance anxiety, improving data naturalness at scale.

Designed the protocol for international replication from day one, rather than treating localization as a later problem, reducing rework for future study phases.

Broader engagement at Apple

Study 1, Lead moderator

This case study. Protocol author, lead moderator, moderator trainer.

Study 2, Lead moderator (current)

The project I'm leading now, selected again as lead moderator on the strength of prior execution. Built around specialized audio equipment, with a planned scale of 1,000 participants.

Study 3, Contributing researcher

Contributed to a third study in a supporting research role across a concurrent research stream.

Study 4, Contributing researcher

Contributing in coordination with international teams around the world, spanning 5+ locales and languages. The protocols and materials I developed are subsequently adopted and used by these global teams.

Lessons learned

What I'd carry forward.

A protocol that serves engineering, design, and research at once only works if you build it with all of them in the room from the start, not by bolting requirements on afterward.

At high session volume, consistency is the deliverable: standardized SOPs and trained moderators protect data integrity more than any single well-run session.

Designing the protocol for replication from day one, across phases, locales, and languages, turned a one-off study into infrastructure other teams now reuse.