Retrieval Engineer (Remote — Worldwide)

As a Retrieval Engineer at Cohere, you will be responsible for building the foundation of our Retrieval-Augmented Generation (RAG) ecosystem, ensuring that enterprise organizations can access and process their proprietary data with industry-leading precision. You will optimize our dense and sparse embedding models, develop sophisticated reranking algorithms, and bridge the gap between large-scale vector databases and LLM performance. Your work will directly impact how Cohere’s flagship models interact with external knowledge, making search more semantic, efficient, and scalable for millions of users worldwide.

- Design and implement state-of-the-art retrieval pipelines that power Cohere's RAG and enterprise search products. - Develop and optimize embedding models to improve semantic understanding across diverse languages and domains. - Research and integrate advanced retrieval techniques such as late interaction (ColBERT), query expansion, and document chunking strategies. - Collaborate with the Product and Engineering teams to reduce latency in real-time retrieval and reranking workflows. - Build robust evaluation datasets and leaderboards to benchmark retrieval accuracy against industry standards. - Contribute to open-source initiatives and technical documentation to support the broader developer community using Cohere’s stack.

- Deep expertise in information retrieval (IR) fundamentals, including vector search, BM25, and hybrid systems. - Proven experience training and fine-tuning embedding models and cross-encoders for production environments. - Strong proficiency in Python and familiarity with high-performance frameworks like JAX, PyTorch, or CUDA. - Practical experience with vector databases (e.g., Pinecone, Weaviate, Qdrant) and large-scale indexing techniques like HNSW or IVF. - Hands-on experience with RAG evaluation frameworks such as Ragas, TruLens, or custom benchmarking suites. - Ability to work independently in a distributed, global team across multiple time zones. - A strong track record of shipping production-grade machine learning code at scale.

Equity, benefits, remote.