Machine Learning Engineer, Safety and Customer Care AI

At Lyft, our purpose is to serve and connect. We aim to achieve this by cultivating a work environment where all team members belong and have the opportunity to thrive.

The Safety and Customer Care (SCC) team at Lyft manages over 1.7 million monthly human and AI interactions and serves as Lyft's primary direct touchpoint with riders and drivers. We handle critical infrastructure that powers both human associates and AI agents to make riders and drivers feel safe and comfortable while riding or driving with Lyft, transforming every support interaction into a moment of genuine connection.

Agentic AI is at the center of how we scale that mission. We fine-tune and align open-source models, build AI-powered support agents, and develop end-to-end AI agents for safety case management, systems that reason over complex, high-stakes cases and drive them to resolution. SCC brings together ML, data, backend, and product engineers alongside data scientists and operations partners to transform these systems.

As a Machine Learning Engineer on the SCC team, you will fine-tune and align models and build AI Agents that power how riders and drivers get help. Your work spans the full loop: post-training open-source models for our domain, composing them into multi-step agents, and building the evaluation that proves they are safe to ship in a customer-facing, safety-critical setting.

• Post-train and adapt open-source LLMs for SCC use cases using SFT, LoRA, and preference-tuning methods (RLHF, RLAIF, RLVR).

• Design and build AI-powered support agents and end-to-end agents for safety case management using LangGraph or equivalent agentic frameworks.

• Own the evaluation data flywheel, offline and online, that defines what "good" looks like and build benchmarks for the team to hill-climb.

• Turn interaction feedback into training data and learning signals, closing the data flywheel that continuously improves the models.

Responsibilities:

• Conduct literature review and build post-training framework and lifecycle. Curate and process human and synthetic data for SFT/LoRA/RLHF/RLAIF/RLVR, and iterate on model quality for real support and safety tasks.

• Develop, evaluate, and productionize AI agents, designing tools, state, and control flow in LangGraph (or equivalent) and taking them through the full agent development lifecycle.

• Build and scale evaluation frameworks, golden sets, rubric-based grading, LLM-as-judge where appropriate, and regression testing.

• Ship models and agents into real-time production, with the monitoring and guardrails needed to operate them safely at millions of interactions a month.

• Apply traditional ML (classification, ranking, gradient-boosted trees) where it's the right tool, and partner with product, ops, and data science to scope problems and define success metrics.