Company
Building the infrastructure for high-value knowledge work.
Copula Lab supplies the benchmarks, expert workflow data, and RL environments that frontier labs need to push models into the professional work that matters most.
Mission
The next act of model deployment is expert work.
After coding, the next frontier is whether models can enter high-economic-value, high-barrier professional workflows. Frontier post-training is shifting from static QA, preference labels, and single-turn tasks toward interactive, verifiable, scalable expert-workflow environments.
Copula supplies SFT and RL-environment data to top model labs, builds high-impact benchmarks grounded in frontier deployment needs, and is building a premium expert-workflow data infrastructure for high-value knowledge work.
Our moat compounds two assets: a horizontal expert-workflow generation infrastructure that lowers the marginal cost of entering new industries, and vertical depth in specific sectors that keeps our benchmarks and data ahead of model capability.
Team
Built by people who shipped frontier models.
Li Liang
Founder
Former head of MiniMax’s Agent business, managing a 20+ person team and reporting to the CEO and COO. Joined Tencent’s management-trainee program in 2021 and MiniMax in 2024, becoming business lead within two years. Built MiniMax’s internal Agent Benchmark from zero and drove 5× revenue growth and eight-figure GMV for Agent in early 2026. Peking University.
Jiaren Cai
Technical Cofounder
Former open-source lead and post-training engineer at MiniMax. Built the vibe benchmark from zero, drove the post-training evolution of the M2 series in coding, and led collaborations with leading bench / data labs — Andon Labs’ Vending-Bench, LMArena, Vals AI’s Vals Index, Artificial Analysis, and Terminal-Bench. Led the international open-source launches of M2, M2.1, and M2.5.
Careers
Join at 0 to 1.
Early members define key systems and key business — not pre-scoped execution. You’ll work against frontier labs’ real needs in post-training, agent training, eval, RL environments, and expert data, with competitive cash, early equity, and room to grow into a direction or executive role.
Open roles
Research Scientist
Own a research or data direction — turn fuzzy problems into benchmarks, data strategy, and training-feedback plans, and analyze model failure modes.
View all rolesAgent Training Infra Engineer
Build the data-production pipeline and RL environments — abstract real tasks into trainable, verifiable, reproducible interactive environments.
View all rolesAgent Data Strategy PM
Define cross-industry agent task scenarios with domain experts; design evals, benchmarks, and rubrics; turn business tasks into data products.
View all rolesBD / Enterprise Sales
Open and grow relationships with top model teams and AI labs; turn fuzzy training and data-procurement needs into deliverable, repeatable products.
View all rolesWork with us.
Whether you’re a frontier lab, a domain expert, or a candidate — we’d like to hear from you.
Talk to us