Lilt | AI Language Expert Project: ESPL (Expert Scaled Prompt & Language) – Pilot Phase
Provided multimodal LLM linguistic evaluation and training contributions for Russian language Speaking and Listening tasks during a pilot phase. Performed preference triad ranking and created error rewrites to generate high-quality golden responses with accompanying rationales in English for engineering feedback. Supported development and refinement of linguistic evaluation guidelines for multimodal model performance. • Evaluated AI outputs in Speaking and Listening categories • Ranked preference triads (HHH) and produced golden responses • Wrote detailed English rationales for linguistic adjustments • Used OneForma workflows for pilot training and assessment