Role: Member of Technical Staff - Research Engineer, Omni

Team: Omni (Multimodal), Research & Engineering

Location: Boston

Location Type: Hybrid

Employment Type: Full-time

About Liquid AI

Spun out of MIT CSAIL, we build general-purpose AI systems that run efficiently across deployment targets, from data center accelerators to on-device hardware, ensuring low latency, minimal memory usage, privacy, and reliability. We partner with enterprises across consumer electronics, automotive, life sciences, and financial services. We are scaling rapidly and need exceptional people to help us get there.

The Opportunity

Omni is where Liquid's vision and audio work converges into a single multimodal model, and the next directions, multimodal MoE and token-efficient encoders, are open problems. This is an end-to-end model training role: you will run open research, data, pre- and post-training, evals, and the inference and deployment path for Omni, working directly with the Omni lead and the vision and audio teams, with our Chief Science Officer close to the research. The team is small and has just shipped a release, so there is no layer between you and the model, and the training infrastructure is being consolidated now rather than inherited.

We are open on level: experienced engineers and strong early-career candidates are both considered.

What We're Looking For

We need someone who:

Owns a task through to done: Takes a loosely defined problem, decides what good looks like, and finishes it well without being chased.

Research curiosity: Goes deep on why a method works and asks about directions the team has not taken yet.

Cares about infrastructure quality: Reviews and hardens code before it lands, including code an AI wrote, rather than treating that as someone else's job.

The Work

Run multimodal training experiments end to end, pre- and post-training, and fold the results into the next Omni release.

Contribute to open research directions such as multimodal MoE and token efficiency for multimodal encoders.

Share the multimodal data work: pipelines, curation, and the data decisions that feed each release.

Take research-grade experiment code and harden it into maintainable infrastructure in the team repo.

Help consolidate Omni's training infrastructure after the latest release: one codebase, one library, and distributed-training choices made and defended.

Own inference and deployment for Omni models, and work across the vision and audio teams on the connective tissue between data, infra, training, and evals.

Experience & Capabilities

What you'll need

Hands-on experience training ML models in at least one non-text modality (vision is the most likely path; audio is rare and can be learned here).

A foundational understanding of how multimodal training actually works, below the research layer, and the ability to explain it from your own past work.

Comfortable sitting between research and engineering: you run experiments and harden the code behind them.

Able to take training infrastructure and inference from experiment code to something maintained and production-ready, and willing to own that path.

Where we would most like to see unusual strength: enjoying the infrastructure and inference side, having consolidated or rebuilt a training stack before, and arriving with your own opinions on libraries and distributed paradigms.

Helpful, not required

Audio modality and speech modeling experience.

Prior inference or deployment experience, including on-device inference.

Exposure to multimodal MoE or encoder token-efficiency research.

A publication record or academic research background; what matters is whether you understand what you built.

What Success Looks Like (Year One)

By the end of your first year, you run Omni model training end to end, including its infrastructure, inference, and half of its data work, and the team's research bandwidth has grown because of it.

You have driven a new capability, such as multimodal MoE, into a shipped release.

The post-release infrastructure consolidation is finished and trusted by both the vision and audio teams.