Research - Synthetic Data

At Mixedbread, we believe that search is the ultimate bottleneck for AI. Our mission is to create the next generation of Information Retrieval systems: truly multimodal, transformer-native, and capable of surfacing just the right context across any format.

We are building a full-stack retrieval research lab that rethinks how search works in the era of large models. Our research spans late interaction methods, multimodal retrievers, and LLM-augmented search, with the goal of pushing retrieval beyond its current limits. Part of building retrieval for all imaginable use-cases is ensuring that we have sufficient data that will allow the model to know how to represent all kinds of information, in all kinds of scenario.

Synthetic data is increasingly becoming one of the core components of all efficient models. Recent efforts have shown that good synthetic data is a lot more than a one-and-done model call. Understanding how to build the best pipelines for retrieval is yet to be understood: if you would like to be the person breaking new grounds in this area, this role is for you.

• Design synthetic data generation pipelines across modalities
• Lead and contribute to applied research projects in Information Retrieval
• Advance Omni, our core search platform, improving retrieval performance and efficiency
• Create innovative feedback loop to constantly improve data generation​
• Collaborate across engineering and product teams to identify key areas of improvements in our models' performance and design the data necessary to bridge that gap.
• Publish and share results through papers, blog posts, and conferences

• Late Interaction:
  Designing the next generation of fine-grained retrieval models beyond ColBERT
• Omni-Modality:
  Building retrievers that unify text, image, and beyond
• Data Refinement:
  Figuring out how to improve existing training datasets, modifying and filtering them to improve both training efficiency and downstream performance.
• End-to-End Synthetic Pipelines:
  Understanding every part of the pipeline, from generation, to quality scoring, to re-writing and curating, among others. Define new components as needed.

• Experience in information retrieval, embeddings, or related fields.
• A deep interest in synthetic data pipelines and data curation work.
• A degree is appreciated but not mandatory for this position.
• Proficiency in Python and ability to implement models from scratch
• Passion for advancing search beyond single-vector similarity methods
• Strong written and verbal communication skills

• Publications in top-tier conferences (NeurIPS, ICML, SIGIR, ACL, etc.) or high-quality blog posts.
• Open-source contributions in Data, IR, NLP, or ML libraries.
• Familiarity with retrieval and late interaction systems.

• Competitive compensation + equity
• Comprehensive health, dental, and vision coverage
• Access to the best AI tools and subscriptions
• Team off-sites + conference attendance
• Transportation support
• Wellness support, including gym membership and sports club subscriptions
• Food support

Mixedbread is an equal opportunity employer committed to building a diverse and inclusive team. We do not discriminate on the basis of race, religion, color, national origin, gender, sexual orientation, age, marital status, veteran status, or disability status.