The opportunity
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.
What you'll do
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.
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.
What they're looking for
- 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.
- + years of industry experience in applied ML/AI, inclusive of an MS or PhD in: Computer Science, Machine Learning, Artificial Intelligence or a related technical field.
