Multimodal Engineer
about the team
about the role
Turn raw video, audio, and sensor recordings into aligned, validated, searchable datasets for EQ's research into people and their interactions. Own the processing and quality pipeline, from ingestion and synchronization to storage, recovery, and research-ready exports.
what you'll own
- Build reliable, resumable pipelines from capture-site drives to cloud storage, with versioned datasets, manifests, and source lineage.
- Build decoding, synchronization, calibration, normalization, and metadata pipelines for multicamera video, multichannel audio, and sensor recordings.
- Implement QA/QC for corrupt or missing files, dropped frames, audio alignment, synchronization drift, and annotation consistency. Recover failed jobs without losing or duplicating data.
- Own object storage, databases, catalogs, indexes, and access controls, including lifecycle policies, cost tracking, and tested restores.
- Run existing models for captioning, tracking, embeddings, and automated quality checks. Make derived features searchable, measure errors, and route uncertain outputs for review.
- Deliver research-ready exports and operate the platform through monitoring, incident response, documentation, and improvements to processing throughput and cost.
what we're looking for
- Five or more years building data infrastructure, including production datasets at tens to hundreds of terabytes or greater.
- Strong Python and SQL, with experience processing large video, audio, or sensor datasets and understanding timing, formats, and data integrity.
- Deep experience with object storage, production data systems, and reliable batch or streaming pipelines, including recovery from real failures.
- Experience with batch or GPU inference, dataset versioning, source lineage, and search or vector indexes.
- Sound judgment about schemas, idempotency, access controls, cost, and production reliability.
Experience with camera calibration, signal processing, FFmpeg/GStreamer, multicamera synchronization, or human-subjects data governance is useful. A PhD is not required. Work with the VLM Researcher on dataset specifications and model outputs; that role owns model research and training.
location and compensation
Available full time or on contract, in San Francisco or remotely. Pay within these ranges depends on location, job-related knowledge, skills, and experience.
- Contract: $50–$100 per hour
- Full time: $120,000–$210,000 per year, plus equity
thank you. your application is in our system.