Applied ML Engineer - Fine-Tuning Open Models & Tools

Vidman AI; United States; Remote; USD 150,000 - 230,000; Full time

Job summary

Vidman AI is seeking an ML engineer to help customers fine-tune open language and vision-language models, from data inspection to evaluating and serving trained weights. You will lead fine-tuning work, build tools for dataset checks and training orchestration, and troubleshoot issues like unstable loss and memory limits, while translating customer needs into product improvements.

The role blends hands-on ML with building internal tooling to make training workflows reliable and scalable, with

Qualifications

• Five+ years of ML production experience.
• Experience fine-tuning LLMs or vision-language models and interpreting training curves.
• Strong Python and PyTorch skills; manage GPU memory.
• Data-first approach; inspect datasets before selecting models.
• Clear communication to explain failed runs to customers.

Responsibilities

• Lead fine-tuning work from dataset review and method choice through run configuration, evaluation, and handoff.
• Build tools for dataset checks, training orchestration, and model evaluation across a large open-model catalog.
• Investigate training and deployment issues, including unstable loss, adapter behavior, tokenization edge cases, and GPU memory limits.
• Identify recurring customer needs and propose prioritized improvements to product and research teams.
• Help customers deploy fine-tuned model weights through the inference stack.

Skills

Python

PyTorch

Data inspection

Communication

Job description

Vidman AI is seeking an ML engineer to help customers fine-tune open language and vision-language models, from data inspection to evaluating and serving trained weights. You will lead fine-tuning work, build tools for dataset checks and training orchestration, and troubleshoot issues like unstable loss and memory limits, while translating customer needs into product improvements.

The role blends hands-on ML with building internal tooling to make training workflows reliable and scalable, with