Job Description:
• Learn, maintain, and improve the synthetic data pipeline and automated training system
• Take ownership of a model and improve it with an agent loop that generates its own data
• Build closed loops where agents train detectors and classifiers, test them against validation sets, and improve them with minimal human input
• Design systems that handle field false positives, flag uncertain detections, and automatically fix the rest
• Lead research on vision-language models, fine-tune them for specific tasks, and use agents to track current techniques
• Solve detection and classification problems for customers and partners by building new models or extending existing ones
• Apply agents to QA automation and code bases that find and fix their own bugs
• Share successful approaches with the team and present results to customers when needed
Requirements:
• Advanced English (C1) proficiency, with the ability to communicate fluently and handle complex, fast-paced technical and professional tasks
• Experience building personal agent harnesses or agentic workflows, with demonstrable work
• Experience running many agents simultaneously, including handling inter-agent communication, token use, and task completion verification
• Professional software engineering experience on both Windows and Linux
• Experience with Git and an issue tracker
• Experience fine-tuning or quantizing open-source models, such as training LoRAs
• Experience running open-source models locally
• Knowledge of the model landscape from early GPT models to current open-source models, including their strengths and limitations
• Clear communication with team members and customers
• Passion for AI and agents, and willingness to experiment with new tools and ideas
• English-only resume/CV submission
• Applicants from every background and country welcome
Benefits:
• Fully remote work
• Access to Claude, Codex, and self-hosted models every day with no limits on applying them to work
• Opportunity to work on models already running in production on edge boxes around the world
• Direct collaboration with the whole team and influence over company-wide AI workflows
• Collaboration with colleagues interested in agent behaviour, harness design, and model quirks