About the role
## About the role Humyn Labs is hiring a Senior Data Engineer to build the pipelines that turn raw multimodal capture — audio, egocentric video and sensor streams from 20+ countries — into versioned, quality-checked, client-ready datasets for frontier AI and robotics teams. **Location:** Bengaluru, Karnataka, India · **Employment:** full-time, on-site · **Experience:** 5+ years ## About Humyn Labs Humyn Labs builds the intelligence layer for physical-world AI. It converts human action — sound, sight, movement and touch — into training-grade multimodal data, captured across 20+ countries in India, Southeast Asia, Latin America and the Middle East: the real environments where physical AI deploys, not the labs where it is built. Two products sit on that capture network — structured voice datasets for frontier teams building speech-to-text, speech-to-speech and multimodal models, and an egocentric video corpus shipped to robotics and world-model teams with labels such as 6-DoF head pose, 3D hand keypoints, depth and object tracks. The data is not just collected; it is evaluated, defended and production-ready. ## What you will do - Design and build batch and streaming pipelines that ingest raw audio, video and sensor (IMU) data from the global capture network into cloud storage. - Build processing stages for unstructured media: transcoding, segmentation, metadata extraction, de-duplication and joins with annotation and contributor data. - Own data quality: automated validation, QC gates and anomaly detection so that bad captures are caught before they reach a client. - Build dataset versioning, lineage and manifests so any delivered dataset can be reproduced exactly. - Package and deliver datasets in the formats client teams ask for (Parquet, WebDataset, Hugging Face datasets, MCAP and similar), on time and to specification. - Orchestrate workflows (Airflow, Dagster or Prefect) and keep pipelines observable, idempotent and cost-efficient. - Work closely with research, annotation operations and infrastructure engineers; mentor junior engineers. ## What you need - 5+ years of data engineering, building production pipelines end to end - Strong Python and SQL - Distributed processing at scale — Spark, Ray, Beam or Dask - Workflow orchestration (Airflow, Dagster or Prefect) and cloud object storage on AWS or GCP - Has worked with large unstructured or media data (audio, video, images, sensor logs), not only tabular warehouse data ## Nice to have - FFmpeg/SoX, audio or video processing libraries. - Experience building datasets for ML training at an AI or data-labelling company. - Lakehouse formats (Delta, Iceberg) and columnar storage design. - Kubernetes and infrastructure-as-code. ## Working model - Full-time, on-site in Bengaluru.
Skills: Python, SQL, Spark, Ray, Airflow, Dagster, AWS, GCP, Data quality, Media pipelines, FFmpeg, Parquet, Dataset versioning