Lead AI Research Engineer

FUNDA is building AI-native research systems for professional users. The Lead AI Research Engineer will establish the scientific and technical foundations for evaluating FUNDA’s LLM and agent systems, define evaluation methods, and use empirical findings to guide technical and architectural decisions.

Responsibilities:

• Define FUNDA’s evaluation strategy, technical architecture, and research methodology for production LLM and agent systems
• Design evaluation frameworks that faithfully represent real agent behavior, including multi-step reasoning, tool use, information retrieval, long-running execution, and generated artifacts
• Develop benchmarks that measure not only final-answer quality, but also reasoning trajectories, evidence use, task completion, robustness, and operational reliability
• Create advanced evaluation methods combining deterministic verification, task-specific grading, process supervision, model-based evaluation, statistical analysis, and targeted expert review
• Lead controlled experiments and comparative studies across models, prompts, tools, data strategies, and agent architectures
• Establish rigorous standards for dataset construction, baseline selection, error taxonomy, experiment reproducibility, statistical validity, and regression detection
• Diagnose complex system failures and distinguish whether they originate from model capability, context construction, tool behavior, data quality, orchestration logic, or infrastructure
• Develop methods for identifying benchmark contamination, evaluator bias, unstable metrics, and differences between offline evaluation and production performance
• Translate research findings into architectural recommendations, release criteria, and prioritised engineering investments
• Build reusable evaluation infrastructure capable of supporting reproducible experiments across different execution environments and agent configurations
• Set technical standards and mentor engineers in evaluation-driven AI development, experimental reasoning, and evidence-based decision-making
• Track relevant advances in LLM evaluation, agent benchmarking, AI reliability, and model supervision, and determine which methods are suitable for production adoption