Workplace type:
Remote / Hybrid
Employment type:
Contract (6 months, extension possible based on performance and business needs)
Seniority:
Mid-Senior level (3–6 years)
Reports to:
Chief AI Officer
Description:
We are seeking a
Foundation Model Training Specialist to support the development and continuous improvement of next-generation video, audio, and multimodal AI models.
You will serve as a critical human-intelligence layer within the AI training pipeline, turning model outputs into structured learning signals. The role combines AI evaluation, human feedback generation, benchmarking, data-quality assessment, and multimodal content analysis to improve model performance, reliability, and user experience.
Reporting directly to the Chief AI Officer, you'll work closely with AI researchers, ML engineers, data scientists, and product teams to accelerate model-quality improvements and support production-scale AI systems.
What you'll do
• Evaluate outputs from video, audio, speech, music, and multimodal foundation models
• Generate structured human feedback for supervised fine-tuning and preference-based training (RLHF)
• Rank and compare model outputs to build high-quality training datasets
• Identify model weaknesses, failure modes, hallucinations, and quality regressions
• Support reinforcement learning and preference-optimization workflows
• Create evaluation rubrics and scoring methodologies; validate annotation quality
• Produce model-quality reports, track improvements across model versions, and present findings to AI leadership
Video evaluation — what you'll assess
Prompt adherence · visual quality · motion realism · temporal consistency · camera-movement quality · human-anatomy accuracy · object permanence · scene continuity · lip synchronization · narrative coherence
Audio evaluation — what you'll assess
Speech naturalness · pronunciation accuracy · emotional expression · accent quality · prosody · clarity and intelligibility · background artifacts · music-composition quality · audio synchronization
Multimodal evaluation — what you'll assess
Video-to-audio synchronization · speech-to-video alignment · prompt-to-output consistency · cross-modal reasoning quality · character and scene continuity · overall multimodal user experience
What we're looking for
• 3–6 years in one or more of: AI operations, AI training, ML support, data-annotation programs, QA, video production, audio engineering, or digital media analysis
• Understanding of generative AI systems, including LLMs, video generation models, speech synthesis systems, text-to-audio models, and multimodal AI
• Ability to evaluate outputs using structured scoring frameworks
• Strong analytical and pattern-recognition skills
Tools & platforms (preferred)
Python · Jupyter Notebooks · Label Studio · Scale AI · Surge AI · Appen · Excel / Google Sheets · AI evaluation platforms