What researchers say
about Corvex

THE CHALLENGE

Ahmad had finished his PhD and moved into independent research before his next institutional position started. His university compute allocation had ended, and cloud costs were unsustainable for the scale of processing his work required — particularly for multi-epoch training runs that could not be interrupted mid-way.

THE APPROACH

Ahmad joined Corvex on a quarterly membership. The onboarding session helped him adapt his workflow to the shared environment within two days. Using the advance schedule, he planned his longer training runs around confirmed compute windows rather than hoping capacity would be available.

THE OUTCOME

Over one quarter, Ahmad completed the processing phase of his research project that had been stalled for three months. He renewed membership for a second quarter and has since contributed two workflow notes to the community documentation library.

THE CHALLENGE

Tan had theoretical knowledge of GPU compute from his coursework but had never configured or run workloads on shared infrastructure. His supervisor needed results from a large-scale experiment within a semester, and Tan did not have time to learn by trial and error across multiple failed runs.

THE APPROACH

Tan enrolled in the Guided Onboarding Programme before joining as a member. Over three sessions across ten days, he worked through environment configuration, job scheduling, and shared-resource norms. He received a starter kit before the first session and took away a practices guide at the end.

THE OUTCOME

Tan's first production run completed successfully within his first two weeks as a member. His supervisor noted that the results arrived a full month ahead of their original estimate. Tan attributed this to not losing time on environment problems that the onboarding had already covered.

THE CHALLENGE

A research unit with five active teams had acquired dedicated GPU hardware but had no system for allocating it fairly across teams. Senior researchers were getting informal priority and junior members were frustrated. The unit head wanted a documented, equitable system before the situation damaged team relationships further.

THE APPROACH

Corvex worked alongside the unit over three months as an independent advisor. The engagement covered scheduling system design, documentation standards, and a community norms framework that the teams themselves reviewed and contributed to. Monthly check-ins kept the engagement on track with written notes after each session.

THE OUTCOME

The unit launched their own fair-access schedule at the end of the engagement. The setup playbook provided by Corvex became the operational reference for the unit's compute coordinator. Six months after the engagement ended, the unit head reported that the scheduling system was still running as designed.