Research Engineer, Post-Training & Vision-Language Models

Cohesity is a leader in AI-powered data security and management. Aided by an extensive ecosystem of partners, Cohesity makes it easy to secure, protect, manage, and get value from data — across the data center, edge, and cloud. Cohesity helps organizations defend against cybersecurity threats with comprehensive data security and management capabilities, including immutable backup snapshots, AI-based threat detection, monitoring for malicious behavior, and rapid recovery at scale.
We’ve been named a Leader by multiple analyst firms and have been globally recognized for Innovation, Product Strength, and Simplicity in Design.
Join us on our mission to shape the future of our industry.

Research Engineer, Post-Training & Vision-Language Models

Want to join the leader in AI-powered document understanding? 

About the Role :

Enterprise documents such as contracts, invoices, filings and reports hold some of the world’s most valuable data, and most of it is still locked in layouts, tables and charts that software struggles to read. The Gaia team at Cohesity is building a world-class agentic document processing product to change that. 

At its heart are vision language models (VLMs) that read documents as images, extract structured data, understand complex layouts, tables and charts, and answer questions about their content. We’re looking for a Research Engineer to push that capability forward. 

You’ll work where research meets production, owning the fine-tuning, evaluation and data pipelines that determine how well our models understand and act on real enterprise data. You’ll have broad scope, significant compute, and a direct line from your work to a shipping product. The problems are open-ended: the path is ambiguous, the signal is noisy, and the right answer takes both research taste and engineering execution. 

Cohesity is the leader in AI-powered data protection and security, trusted by over 13,600 enterprise customers, including more than 85 of the Fortune 100 and nearly 70% of the Global 500. Together they entrust us with over 200 exabytes of data. We’re backed by NVIDIA, IBM, HPE, Cisco, AWS, Google Cloud and others, and headquartered in Santa Clara, CA, with offices around the globe. 

Want to join a scaled multi-billion ARR private company? On a team that’s operating like a Series A company?

HOW YOU'LL SPEND YOUR TIME HERE:

• Evaluate, select, and adapt VLMs (open-weight and API-based) for document parsing, layout understanding, table and chart extraction, OCR-heavy content, and document question answering. 

• Fine-tune vision-language models using SFT and latest RL techniques such as RLHF / RLAIF / RLVR, GRPO or PPO-style RL for document understanding to hill climb towards the best document understanding in the industry. 

• Build reinforcement learning (RL) training platforms, design reward signals, and optimize model reasoning. 

• Build and maintain evaluation harnesses to measure model performance across capability and reliability dimensions. 

• Create pipelines for synthetic data generation to support training and evaluation. 

• Develop and refine post-training recipes that measurably improve production model quality. 

• Collaborate with the agentic engineering team so agents can perceive and act on visual inputs such as documents, screens, and diagrams. 

• Debug training instabilities, regressions, and unexpected model behavior; establish best practices for reproducible training and experiment tracking. 

WE’D LOVE TO TALK TO YOU IF YOU HAVE MANY OF THE FOLLOWING:

Experience

• 5+ years of professional experience in machine learning engineering or research. 

• Hands-on experience training, fine-tuning, using RL techniques or evaluating large language or vision-language models. 

• Track record of taking models from experimentation to production. 

Education

• Master’s degree or PhD in Computer Science, Machine Learning, or a related field preferred. 

Required Skills

• Languages: Python 

• ML/AI: PyTorch (or JAX), Hugging Face ecosystem, supervised fine-tuning, VLMs/LLMs 

• Evaluation: building eval harnesses and benchmarks for model quality 

• Data: synthetic data generation, data curation pipelines 

Preferred / Nice-to-Have Skills

• Document AI experience: OCR, layout analysis, table and chart extraction 

• Visual grounding and region-level reasoning 

• Post-training and distillation for smaller or on-prem-deployable models 

• Publications or open-source contributions in post-training, alignment, or multimodal models 

• Prior experience in enterprise data protection, security, or infrastructure software 

Expectations from the Role

• Balance research exploration with engineering rigor and reliable delivery. 

• Work effectively across research and engineering and stay productive amid ambiguity and shifting priorities. 

• Communicate experimental results and tradeoffs clearly to technical and non-technical stakeholders. 

• Be hands-on: write and review production-quality code and run experiments regularly. 

Disclosure Pursuant to Applicable State Equal Pay Transparency Laws - This position has a starting pay range as listed below. Actual salary depends upon many factors, including a candidate’s skills, qualifications and experience, location, and salary expectations, and therefore a starting salary at the low end, high end, or even above the stated range may be offered. This position may also be eligible for bonus compensation, commission (if in a sales function), and/or equity grants. Additionally, full-time employees are eligible to participate in our comprehensive benefits framework, including health and wellness benefits, vacation, paid holidays and refresh days, 401(k) retirement plan, life and disability insurance coverages, and other benefits the Company may offer from time to time.

$255,600.00-$284,000.00

The compensation noted above is based on an annualized hourly rate assuming normal full-time employment.

We strongly prefer candidates who are currently located in or near the designated job location. Candidates outside the area should apply only if they are committed to relocating prior to their start date and have the legal right to work in the job location.