Careers

ML Engineer

EngineeringRemote-friendly

Build the speech, diarization, and transcript intelligence systems that power DraftTheRecord.

About the role

DraftTheRecord is building AI products for court reporters, transcribers, and legal teams who need fast, accurate, and trustworthy transcripts. As an ML Engineer, you will work on the systems that turn messy legal audio into structured, speaker-aware transcripts that professionals can rely on.

This role sits close to the product. You will prototype models, evaluate transcription quality, improve diarization and formatting behavior, and ship ML-powered features that directly affect customer workflows.

Responsibilities

  • Improve speech recognition, speaker diarization, punctuation, formatting, and transcript post-processing pipelines.
  • Design evaluation sets and metrics for legal audio quality, including speaker identification, quiet audio, overlapping speech, names, and court-specific formatting.
  • Build production-ready model serving and batch processing workflows in partnership with backend and full-stack engineers.
  • Analyze customer audio and transcript quality issues, identify root causes, and turn findings into durable model or product improvements.
  • Experiment with model prompting, fine-tuning, retrieval, and post-processing approaches for legal transcription use cases.
  • Create internal tools that make model evaluation, regression testing, and data review faster and more reliable.

What we're looking for

  • Strong Python experience and practical experience shipping ML systems beyond notebooks.
  • Experience with speech, audio, NLP, LLMs, or document intelligence systems.
  • Comfort designing rigorous evaluations and debugging model behavior with real-world data.
  • Ability to work across model quality, infrastructure, and product constraints.
  • Clear communication and strong ownership in ambiguous problem spaces.

Nice to have

  • Experience with ASR, diarization, forced alignment, VAD, or audio embeddings.
  • Experience with legal, healthcare, finance, or other high-accuracy regulated workflows.
  • Familiarity with cloud ML infrastructure, queues, containers, and observability.

What success looks like

  • Customers see measurable improvements in transcript quality and speaker identification.
  • The team can catch model regressions before they reach users.
  • ML improvements ship as reliable product behavior, not one-off experiments.