Sandra Kublik is an AI educator, author, and customer-education specialist at Cohere who helps enterprises turn language models into useful, trustworthy working tools. Her career spans developer advocacy, retrieval-augmented generation, open-weight multilingual models, and the practical challenges of enterprise AI adoption.
Before Cohere, Kublik advocated for machine-learning teams at Neptune.ai and co-founded Nextgrid.ai and Deep Learning Labs. In 2022, she co-authored a practical book on building products with GPT-3 with Shubham Saboo, introducing developers to applications built with the OpenAI API.
At Cohere, she moved from developer relations and community education into collaborative model research and customer adoption. Her writing about an AI Tinkerers hackathon documents how an independent team applied Cohere Rerank; she also co-authored the Aya Expanse technical report on open-weight multilingual models. More recently, she joined Cohere’s customer-education team, helping organizations identify productive workflows and build durable confidence in enterprise AI.
- Retrieval that users can verify. Effective retrieval systems must separate external documents from conversation history, overcome models’ bias toward information appearing early in documents, and recognize when newly retrieved material should supersede pretrained knowledge. Fine-grained citations let users inspect the evidence behind consequential answers.
- Open-weight models with practical infrastructure. Her work explaining Command R and Command R+ connects model capabilities to usable application architecture: document uploads, configurable retrieval, Next.js interfaces, SQL-backed conversation history, and flexible deployment.
- Multistep tool use with error recovery. Useful enterprise systems can retrieve documents, compare findings, generate summaries, and complete follow-up actions while adjusting their plans when tools fail. Making those actions traceable strengthens accountability.
- Multilingual access and sustained adoption. Through Aya Expanse, Kublik contributed to a broader effort involving multilingual preference training, model merging, and cross-language evaluation. Her customer-education work addresses the complementary organizational challenge: turning initial experimentation into informed, repeatable use.
