Founding AI/ML Engineer, Authorization

Cephas Consultancy Services Private Limited

United States, California, San Francisco

Headcount: 1

Job Details

  • Experience: 2–5 years
  • Salary: Not specified
  • Equity: Competitive equity
  • Visa sponsorship: H-1B, O-1
  • Employment type: Full-time
  • Work type: In person

Benefits & Perks

  • Meals
  • Health, dental, and vision insurance

About This Role

As the Founding AI/ML Engineer, Authorization, you will lead how we decide whether an agent is allowed to do what it is about to do. A person approves an intent (“book me a flight to Austin under $400”), and the agent takes dozens of actions. Your models will decide at each step whether an action still matches what was approved, how risky it is, and when to stop and ask. The role starts as an individual contributor leading the modeling, with room to build a team. This is an in-person role at Lane’s San Francisco office, roughly six days a week (Sunday through Friday), with flexibility on weekends.

You will directly

  • Own the data pipeline and modeling design for our two main models, including a small language model focused on product and cart context.
  • Turn plain-language approvals into checkable authorization envelopes covering merchants, amounts, items, substitutions, and time windows.
  • Score actions to catch drift, substitutions, prompt injection, and overspending before a payment goes through.
  • Make risk decisions using agent identity, user history, device, and merchant signals: approve, step up, or refuse.
  • Own evaluations, labeling, and metrics, tracking false approvals and false blocks as product metrics.
  • Ship models that run in the purchase path under tight latency and cost budgets.

Requirements

  • Hands-on experience training ML or language models, including at least one system that made real decisions in production.
  • Python and TypeScript.
  • Sound judgment about LLMs: where they are reliable, where they fail, and how to measure them.
  • Ability to ship your own models and design evaluations before building models.
  • Depth in at least one of these areas: fraud or risk modeling, policy or authorization systems, or LLM evaluation and guardrails.
  • Willingness to work a roughly six-day week (Sunday through Friday).