Photo by Samsung Memory on Unsplash
Project Overview: Redesigning Data Flow Around Hardware Limits
At Meta, I conducted a multi-session, repeat-measures speech data collection using an early AR glasses prototype across a 3×5 within-subjects experiment (roughly 450 trials per participant). To capture continuous sensor streams without hitting onboard storage limits, we streamed data to a host machine via a cable—deliberately using the device outside its intended untethered operating mode. This introduced severe physical strain on a fragile multi-pin connector, causing frequent disconnects that originally wiped out entire 90-minute sessions. I inherited this setup across two iterations to make it survive contact with real participants and unreliable hardware.
My Role
I owned the end-to-end execution and operational survival of the project across both iterations. My responsibilities included:
- Managing participant recruitment and session execution.
- Diagnosing hardware vulnerabilities and negotiating hardware/software fixes with research scientists and lab managers.
- Revisiting and revising protocols, documentation, and operational workflows.
- Training and onboarding the incoming Research Assistant for iteration two.
Methodology & Execution: Triage, Redesign, and Hand-Off
I employed a rigorous, two-phase diagnostic and engineering approach to transform an unscalable setup into a repeatable process:
- Iteration One (Field Triage & Root Cause Analysis): Specified longer extension cables to reach the desktop lab machine, which unfortunately introduced weight and mechanical stress. Collaborated with the lab manager to devise a mechanical frame fix and layered on participant-side mitigations (routing cables behind the back or holding slack in their lap) to reduce wear.
- Software-Driven Risk Containment: Challenged the all-or-nothing data architecture where 15 sections ran as a single continuous recording. Successfully negotiated with the research scientist to segment recording files into natural break points, converting total session loss into isolated, manageable data blocks.
- Objective Hardware Assessment: Delivered an honest hardware evaluation advising leadership that the prototype was unsuited for large-scale tethered collections without permanent mechanical reinforcement.
- Iteration Two (Process Hand-Off & Standardization): Standardized lessons learned by implementing 20-minute time-capped windows and flexible “complete as many as possible” targets. Conducted dual pilot testing and authored comprehensive documentation to equip the incoming RA for independent success.
Key Findings: Lessons in Hardware and Data Architecture
This project yielded critical technical and operational insights that reshaped how our team viewed prototyping constraints:
- Using hardware outside design intent has a cost: Streaming to a host avoided onboard storage ceilings but compromised connector reliability. Recognizing and planning for this trade explicitly is essential.
- Hardware fixes can create compounding problems: The longer cable solved reach limitations but worsened connector strain. Hardware mitigations must always be evaluated as an interconnected system.
- Failure severity is a software decision: The connector was bound to fail intermittently, but whether that cost 3 minutes or 90 minutes was entirely determined by data segmentation.
- Wear compounds over time: Early-session feasibility metrics were artificially optimistic because failure rates escalated cumulatively as hardware experienced prolonged use.
- Reframing the problem: Treating an unreliable connection as a data architecture problem to be contained (rather than just a physical hardware flaw to be fixed) fundamentally changed our resilience.
Outcome & Strategic Impact: Eliminating Data Loss & Informing Strategy
This research directly optimized our operational workflow and influenced team-wide hardware strategy:
- Eliminated Catastrophic Data Loss: Software segmentation and time-capped protocols completely removed the all-or-nothing failure mode, dramatically increasing overall data retention.
- Informed Hardware Strategy: My candid hardware assessment provided leadership with a realistic evaluation of the AR prototype’s true operational limits for future large-scale collections.
- Seamless Operational Continuity: Iteration two launched with a fully tested protocol, robust documentation, and a thoroughly prepared RA, preventing the team from having to rediscover the same operational pitfalls.
