AI training data and engineering services
G2i
G2i supplies engineering teams and AI training services to frontier AI labs and software companies. Its senior engineers turn software workflows into evaluation tasks, scoring rubrics, training data, and reinforcement-learning environments. These environments reproduce interfaces, API behavior, edge cases, and success criteria so labs can train and evaluate agents on realistic engineering work. Expert reviewers assess code and model outputs, refining the tasks and feedback used in post-training.
Founded by Gabe Greenberg, who serves as CEO, G2i grew from engineering hiring rooted in the React and JavaScript communities into AI training, beginning frontier-lab RLHF pipelines in 2024. Its research examines the quality of coding-agent evaluations: a TypeScript review of Multi-SWE-bench assessed 210 tasks through independent developer evaluations, consensus review, and senior quality checks. G2i identified 87 tasks with both good specifications and excellent test coverage, demonstrating why benchmark results depend on task quality as well as agent capability.
As of August 2026, the company reported 200 full-time engineers and an 8,000-engineer bench. Customers can engage G2i for individual hires, embedded engineering teams, or managed services in which it takes responsibility for delivery. That service range connects its engineering talent business with the specialist review and environment-building work required by AI labs.
1 talk
Newest first1 speaker at AIE
Affiliations reflect their AIE appearances, not necessarily current employment.
Messages from the stage
Verifiers can fail in both directions
Khial highlights how weak verifiers can accept incorrect implementations while rejecting valid alternatives, undermining the benchmark's assessment of a solution.
Affiliations reflect each recorded session, not necessarily current employment.
