Member of Technical Staff; Data Acquisition

Member of Technical Staff (Data Acquisition) About the Role

Your mission is to build and operate the ingestion systems that turn the open web and large-scale audio sources into reliable, well-structured corpora for training Sanas's frontier speech models. You'll own the machinery that acquires, extracts, filters, versions, and delivers audio data to our training pipelines — and you'll work directly with our research scientists to close the loop between what we collect and how it moves model quality.

• Own and lead engineering projects across the full data acquisition stack — web crawling, audio ingestion, source discovery, and dataset delivery to training pipelines.
• Build and operate large-scale distributed crawling infrastructure capable of continuously discovering and ingesting audio at scale across languages, accents, domains, and recording environments.
• Develop specialized crawlers for high-priority audio sources with source-specific extraction and normalization logic.
• Run experiments to evaluate crawling strategies, extraction methods, and ingestion tradeoffs; analyze results to identify gaps, redundancy, and coverage improvements across speaker demographics and language pairs.
• Build ingestion pipelines that scale reliably across large data campaigns, with automated audio quality filtering — SNR estimation, clipping detection, codec artifact identification — as a first-class pipeline stage.

• Design and deploy highly scalable distributed systems capable of handling petabytes of audio data — from raw acquisition through quality filtering, deduplication, segmentation, and versioned dataset generation.
• Architect and implement indexing and search capabilities over large audio corpora — enabling fast lookup by language, speaker, acoustic condition, duration, and quality tier.
• Build and maintain backend services for data storage, including key-value databases, metadata synchronization, and manifest management across dataset versions.
• Deploy and operate acquisition infrastructure in a Kubernetes / Infrastructure-as-Code environment; perform routine system health checks and respond to production issues quickly.
• Collaborate closely with data processing, architecture, and ML platform teams to ensure smooth data flow from acquisition through to training‑ready outputs.

• Work closely with legal to handle compliance, data privacy, and licensing matters across all acquisition sources — maintaining a clear audit trail of provenance, permitted use, and commercial training rights for every dataset.
• Enforce speaker consent documentation, GDPR requirements, robots.txt and ToS adherence, and audio retention policies across all ingestion pipelines.
• Manage relationships with third‑party data vendors — writing precise acquisition briefs, evaluating quality on delivery, and ensuring sourced data meets Sanas's licensing and quality standards.

• 4+ years of experience in data engineering, ML data infrastructure, or backend systems engineering — with direct experience building large‑scale data ingestion or crawling systems.
• Strong Python and systems engineering skills — you build robust, maintainable infrastructure, not just one‑off scripts.
• Hands‑on experience with distributed systems design: you've built systems that handle failure gracefully, scale horizontally, and recover cleanly.
• Experience with web crawling infrastructure at scale including handling rate limiting, deduplication, and content extraction.
• Proficiency with cloud platforms (AWS or GCP), object storage (S3/GCS), and container orchestration (Kubernetes).
• Comfort working with audio processing tooling — ffmpeg, librosa, torch audio, sox — and experience handling large volumes of audio files.
• Strong data quality instincts: you instrument pipelines, surface issues proactively, and treat data correctness with the same rigor as software correctness.

• Experience building speech or audio datasets for ASR, TTS, speech enhancement, or speaker verification model training.
• Familiarity with major open speech corpora — Common Voice, Libri Speech, Vox Populi, AISHELL — and their sourcing and quality characteristics.
• Experience with data versioning tools.
• Background in multilingual or low‑resource language data collection.
• Experience with annotation and labeling platforms.
• Familiarity with speaker diarization, language identification, or automated audio quality estimation models used for data filtering at scale.