Machine Learning Engineer, Contract Understanding

What you will do

• Own the clause extraction pipeline for deliverables, exclusions, revision caps, assumptions and acceptance criteria across PDF and Word SOWs.
• Build and maintain evaluation sets drawn from consented, anonymised contracts, and report precision and recall per clause type every release.
• Design prompting, retrieval and fine-grained classification that runs on zero-retention inference endpoints, with no customer data used for training.
• Calibrate confidence scores so low-certainty requests route to the PM review queue instead of the alert channel.
• Partner with Design on how evidence, citations and uncertainty are shown to project managers.

What you bring

• 4+ years shipping NLP or document AI systems to production.
• Practical experience with LLM evaluation, structured output and retrieval over long documents.
• Strong Python, and the habit of writing evaluation harnesses before changing a prompt.
• Comfort reading contract language and noticing when a clause is ambiguous.

Why this role matters

Agencies trust TWINJAM with their client channels and contracts because we are careful, quiet and accurate. Everyone on the team protects that trust: we review flags with customers, read real SOWs, and measure ourselves on how rarely a PM has to dismiss an alert.