Multimodal Data Engineer

Abaka AI•Mountain View, CA

2d•Remote

About The Position

Abaka AI is seeking a Multimodal Data Engineer in the United States to develop the systems and workflows necessary to transform complex data requirements into high-quality datasets for advanced AI teams. The role involves end-to-end project ownership, from initial feasibility assessment and quality standard definition to building scalable pipelines, coordinating technical efforts, and ensuring datasets meet specifications. This position covers multiple data modalities and all stages of the data lifecycle, including sourcing, processing, annotation, quality control, storage, and delivery. The successful candidate will join the existing data engineering team, contributing to the enhancement of infrastructure, tooling, and operational standards as Abaka AI expands. This role is ideal for an engineer who thrives on hands-on systems work, takes responsibility for project delivery, and can translate ambiguous client needs into concrete technical plans.

Requirements

3+ years of experience in data engineering or building large-scale data systems, with hands-on ownership of production data workflows.
Experience delivering datasets or data systems end to end, including scope, timeline, quality, cost, and acceptance criteria for external clients or internal model teams.
Hands-on experience with at least two data modalities, such as text, images, audio, video, or 3D point clouds.
Experience designing and operating large-scale data pipelines using distributed processing tools such as Spark, Ray, or Flink; orchestration tools such as Airflow, Dagster, or Argo; cloud object storage; and containerized batch compute.
Working knowledge of the infrastructure behind data pipelines, including cloud permissions, storage layout and lifecycle rules, compute provisioning, orchestration environments, and cost monitoring.
Experience designing data quality and acceptance processes, including sampling plans, defect categories, measurable quality gates, or model-assisted quality control.
Familiarity with data privacy and security practices, including access control, anonymization, license and provenance tracking, and encryption.
Ability to translate ambiguous requirements into actionable technical plans with explicit assumptions, resource needs, risks, and acceptance criteria.
Strong ownership, sound judgment, and comfort working across engineering and operations in a fast-moving environment.

Nice To Haves

Experience preparing pretraining, supervised fine-tuning, RLHF, or evaluation datasets for LLMs or multimodal foundation models.
Experience with data for autonomous driving, robotics, embodied AI, or other sensor-based domains.
Experience with large-scale deduplication, data quality scoring, or dataset composition design.
Experience building annotation platforms or internal data tools, or working with external annotation vendors.
Experience modeling and optimizing the cost of GPU- or compute-intensive data pipelines.
Experience establishing engineering standards and improving workflows on a growing team.

Responsibilities

Assess incoming data requirements for feasibility, technical challenges, risks, cost, and delivery timeline.
Define execution plans, resource needs, quality specifications, and written acceptance criteria before work begins.
Own technical delivery for assigned projects, from scoping through final handoff.
Break requirements into tasks with clear owners, priorities, deadlines, and deliverables, and coordinate work across engineers, annotation teams, and external vendors.
Build and improve scalable pipelines for multimodal data sourcing, processing, cleaning, annotation, quality assurance, storage, and delivery.
Establish quality gates at ingestion and before delivery.
Define measurable checks, identify checks that could not be performed, and prevent data that fails agreed specifications from being delivered.
Improve visibility into project and dataset status through tools such as requirement-to-delivery trackers and dataset indexes covering progress, ownership, yield, inventory, quality, cost, sample links, and delivery history.
Help operate the infrastructure behind our data workflows, including object storage, batch processing, compute and GPU resources, orchestration environments, annotation platforms, and internal data services.
Track infrastructure usage and project costs, and identify ways to improve quality, throughput, and efficiency without compromising delivery commitments.
Standardize recurring workflows, automate manual steps, build internal tooling, and document reusable practices based on project retrospectives.
Partner with client-facing teams and foundation model teams to clarify data needs, communicate technical tradeoffs, and raise feasibility or quality risks early.