AI research engineer

Pathos • San Francisco Bay Area, CA

1d

About The Position

Mission Design, train, ship, iterate on, and innovate on the AI brains behind Pathos’ AI Therapist. Combine research, data science, and engineering to create models, orchestration, and evaluation systems that make therapy conversations deeply effective, clinically grounded, and safe. Outcomes Improve quality of AI Therapy: Deliver measurable improvements in conversation quality, therapeutic alliance, and user outcomes through fine-tuning strategies, training data curation, building RL environments, new model architectures and other AI innovations. Improve evaluation of AI quality: Improve on and maintain a robust eval stack that includes scripted tests, LLM-as-judge evaluations, human ratings, and safety checks. Improve automated regression testing, detection of defects, and observability (eg dashboards). Own AI system. Build, maintain, and iterate on the production codebase that delivers AI therapy and supports the evaluation and iteration of our AI. Productionize Models and Pipelines. Own the path from notebook to production: training jobs, model packaging, deployment, monitoring, and rollback strategies. Keep latency, reliability, and cost within agreed budgets while enabling rapid iteration on new ideas. Improve Safety, Alignment, and Clinical Guardrails Work with clinicians and internal experts to encode clinical guidelines into prompts, reward functions, tools, and filters. Proactively identify and reduce harmful or low-quality behaviors through targeted experiments, red teaming, and mitigations. Own Research Roadmap and Experiment Velocity Run high-quality experiments from hypothesis to analysis to improve our understanding of what matters and what works. Shape and execute a focused R&D roadmap. Collaboration with Clinicians, Product, and Engineering. Translate product and clinical requirements into concrete model and system changes. Partner with full-stack product engineers so that new AI capabilities are easy to integrate and maintain in the product.

Requirements

• LLM and Applied ML Depth. Demonstrates strong experience with large language models, including fine-tuning, training data design, and model selection. Knows how to move core metrics on conversation quality and user outcomes, rather than chasing generic benchmarks. Can look at evals, transcripts, and metrics and quickly form grounded hypotheses for improvement.
• Ships clean, maintainable, quality code . No only do you know how transformers work, but you are also an engineer that has experience shipping production-level code and/or maintaining an AI system in production.
• Data Engineering Skills. Can set up production-level data pipelines for training new models, evals, analysis, etc.
• Scientific Mindset. You formulate hypotheses, and you are good at evaluating them (eg through experiments, data analysis, etc). You are consistently learning at the cutting edge, and you’re able to leverage and communicate those learnings to make the entire company more successful.
• Ruthless Prioritizer. You are keenly aware of how to provide company value and to prioritize projects accordingly. Resistant to nerd-sniping .
• Quality Obsessive: Refuses to ship subpar work, continuously improving the codebase.
• Fast: Prioritizes speed by leveraging AI, breaking down complex tasks, shipping early, optimizing for learnings, iterating quickly, and avoiding over-engineering.
• Strong communicator. You can work collaboratively in a positive way. Sees others perspectives. Strong opinions, loosely held. Focused on user/business value, not ego.

Nice To Haves

• Personal or other experience with therapy or coaching
• Domain knowledge of psychology, neuroscience, therapy, or coaching.

Responsibilities

• Improve quality of AI Therapy: Deliver measurable improvements in conversation quality, therapeutic alliance, and user outcomes through fine-tuning strategies, training data curation, building RL environments, new model architectures and other AI innovations.
• Improve evaluation of AI quality: Improve on and maintain a robust eval stack that includes scripted tests, LLM-as-judge evaluations, human ratings, and safety checks. Improve automated regression testing, detection of defects, and observability (eg dashboards).
• Own AI system. Build, maintain, and iterate on the production codebase that delivers AI therapy and supports the evaluation and iteration of our AI.
• Productionize Models and Pipelines. Own the path from notebook to production: training jobs, model packaging, deployment, monitoring, and rollback strategies. Keep latency, reliability, and cost within agreed budgets while enabling rapid iteration on new ideas.
• Improve Safety, Alignment, and Clinical Guardrails Work with clinicians and internal experts to encode clinical guidelines into prompts, reward functions, tools, and filters. Proactively identify and reduce harmful or low-quality behaviors through targeted experiments, red teaming, and mitigations.
• Own Research Roadmap and Experiment Velocity Run high-quality experiments from hypothesis to analysis to improve our understanding of what matters and what works. Shape and execute a focused R&D roadmap.
• Collaboration with Clinicians, Product, and Engineering. Translate product and clinical requirements into concrete model and system changes. Partner with full-stack product engineers so that new AI capabilities are easy to integrate and maintain in the product.

What This Job Offers

• Job Type: Full-time
• Career Level: Mid Level
• Education Level: No Education Listed