Research Engineer

Vidman AI; United States; Remote; USD 140,000 - 230,000; Full time

Job summary

Vidman AI is advancing the quality and efficiency of open-language models used in production. This role focuses on routing, fine-tuning, and alignment to deliver measurable improvements in quality, latency, and cost.

The candidate will develop evaluation systems, reproduce promising research, and collaborate with product engineers to make research results visible in product interfaces. A strong emphasis on reproducible experimentation and rigorous claims is essential.

Qualifications

• Five+ years in ML research or engineering with large language models.
• Practical PyTorch fine-tuning, distributed training, and PEFT (LoRA/QLoRA).
• Strong evaluation discipline with evidence-based claims.
• Reproducible engineering practices.

Responsibilities

• Develop routing strategies across an open-model pool and measure quality, latency, and cost.
• Design fine-tuning and alignment methods for dedicated GPU environments.
• Build evaluation systems to catch regressions before users are affected.
• Review research, reproduce methods, and move results into practice.
• Collaborate with product engineers to expose results in product interfaces.
• Document findings in internal notes and product documentation.

Skills

Open-weight models

PyTorch

Distributed training

Evaluation discipline

Reproducible experiments

Tools

LoRA/QLoRA

PEFT techniques

Job description

Role overview

Improve the quality and operational behavior of open language models serving production workloads. The work spans model routing, fine-tuning, alignment, and evaluation, with an emphasis on turning research into measurable improvements in quality, latency, and cost.

Responsibilities

• Develop and evaluate strategies for routing requests across an open-model pool, measuring effects on quality, response time, and cost.
• Design fine-tuning and alignment approaches—including supervised fine-tuning, LoRA or QLoRA, preference optimization, and continued pre-training—for dedicated GPU environments.
• Build evaluation systems that identify regressions before they affect users.
• Review research, reproduce promising methods, and move effective results into practice.
• Work with product engineers to make research results visible and controllable in product interfaces.
• Document findings in internal notes, product documentation, and occasional public writing.

Requirements

• At least five years in machine learning research or engineering, including substantial work with large language models.
• Practical experience fine-tuning open-weight models using PyTorch, distributed training, and parameter-efficient techniques.
• Strong evaluation discipline and the ability to demonstrate evidence for claims of model improvement.
• Engineering practices that make experiments reproducible by others.