Senior AI Research Engineer

Phaidra builds AI-powered control systems for the industrial sector that enable facilities to learn and improve over time. The Senior AI Research Engineer will develop the cognitive and behavioral layer of Phaidra’s agents by designing reasoning frameworks, prompts, guardrails, evaluation systems, and feedback loops. The role also partners with engineering, product, domain, and customer-facing teams to improve agent reliability, interpretability, safety, and usefulness.

Responsibilities:

• Design and improve agent reasoning frameworks that help Phaidra’s agents solve complex industrial problems using tools, telemetry, ontology, historical context, and domain logic
• Define and maintain agent personalities, communication patterns, behavioral boundaries, and escalation rules across Phaidra’s agent portfolio
• Develop prompts, system instructions, tool-use policies, and execution loops that help agents act consistently and effectively in industrial contexts
• Build and maintain agent evaluations that measure behavioral quality, domain correctness, user usefulness, safety, hallucination risk, tool-use reliability, and alignment with business requirements
• Translate subject-matter expertise into reusable reasoning patterns, decision frameworks, eval rubrics, and behavioral tests
• Analyze production interactions, user feedback, telemetry, and qualitative signals to identify where agents are succeeding, drifting, or failing
• Partner with Agent Reasoning, Ontology, AI Platform, Product, Solutions, and customer-facing teams to improve agent reliability and usefulness
• Help agents reason over derived tags, operational math, alarms, recommendations, facility context, and customer-specific constraints
• Track state-of-the-art developments in agentic AI, evaluation methods, prompting, tool use, behavioral assurance, and domain-specialized LLM systems
• Contribute to Phaidra’s broader agent strategy by ensuring our agents are not only technically capable, but trustworthy, interpretable, and valuable to users

Requirements:

• Strong applied AI, machine learning, data science, research engineering, or software engineering background
• Experience working with LLMs, agents, copilots, autonomous workflows, or other AI systems that reason over tools and context
• Ability to translate ambiguous product and domain problems into concrete experiments, prototypes, evals, and production improvements
• Experience evaluating model or agent outputs using qualitative and quantitative methods
• Strong product judgment: you understand what makes an AI system actually useful to users, not merely technically impressive
• Excellent written communication skills, with an ability to craft precise instructions, behavioral guidelines, research notes, and evaluation criteria
• Comfortable collaborating across engineering, product, domain experts, and customer-facing teams
• Ability to reason about ambiguity, uncertainty, edge cases, and failure modes in AI systems
• Good working knowledge of Python
• Shares Phaidra's company values: Agency, Velocity, Craft, Truth
• Candidates who are based in the UK or North America (Eastern timezone)
• legally authorized to work in the specified location(s) above
• Candidates requiring visa sponsorship, either now or in the future, are not eligible for hire
• Experience designing or evaluating LLM agents, tool-using AI systems, prompt systems, guardrails, eval harnesses, red-teaming, or behavioral QA
• Experience with HVAC, datacenter operations, EPMS, BMS/BAS, industrial controls, alarm triage, or root-cause analysis
• Experience with time-series analysis, anomaly detection, derived tags, operational math, symbolic reasoning, or model-informed decision-making
• Experience with RAG systems, knowledge graphs, ontologies, semantic retrieval, or context engineering for AI agents
• Experience building dashboards, evaluation datasets, labeling workflows, or feedback loops for AI products
• Familiarity with agent frameworks, MCP/tool interfaces, LLM observability, or production AI infrastructure
• Track record of turning research ideas into reliable product capabilities