Agentic AI Engineer (Computer Vision)

CVEDIA builds computer vision models and applications that run in real time on server and edge hardware around the world. We train our models on synthetic data, and much of our model work already runs through AI agents. We have made the full move to this way of working across the whole company, and we are looking for people who want to help push it further. This job requires Advanced English (C1) proficiency, with the ability to communicate fluently and handle complex, fast-paced technical and professional tasks. The Role We are looking for an engineer to design and build the agentic systems that generate our synthetic data, train our detectors and classifiers, and iteratively improve them: closed-loop pipelines where agents generate data, train models, evaluate results, and decide what to generate next. This role sits at the intersection of agents, computer vision training, and performance engineering. You will not just use AI tools to write code faster. A large part of the work is building the agents themselves. Our agents are how we work now: everyone here builds with them daily, and you will too. What You Will Do Design, build and operate agentic systems, such as multi-agent pipelines, tool-using agents, and evaluation loops, that generate synthetic data at scale and train computer vision models end to end Fine-tune vision-language models as part of these pipelines, including data preparation, training, evaluation and iteration Write the production code these systems run on: Python services, training pipelines, and agent orchestration, with C++ where performance demands it Profile and optimize across the stack: GPU utilization, memory, throughput, and cost. You will own the efficiency of the pipelines you build, not just their output Build the feedback loops that make models better: automated evaluation, error analysis, and targeted data generation driven by model failures Research and prototype: read papers, test new approaches for agentic training and synthetic data, and bring the ones that work into production Work directly with the team lead. In a company this size, your architectural decisions shape how the whole engineering team works What You Bring Deep experience building agentic AI systems in production, such as LLM agents, tool use, multi-agent orchestration, and structured generation. This is the core of the role, not a nice to have Strong computer vision training experience: you have trained detectors, classifiers, or vision models end to end and know how data quality drives model quality. Experience with synthetic data is a strong plus Experience fine-tuning vision-language models such as Qwen-VL, LLaVA, PaliGemma, InternVL, or similar, including LoRA and full fine-tuning workflows Strong software engineering skills in Python, with C++ experience. You write code that is optimized, structured, and built to be maintained, because agents will read and extend it A track record of performance work: profiling, identifying bottlenecks, and optimizing code for speed and GPU efficiency Research literacy: you keep up with the literature on agents, synthetic data, and model training, and can evaluate a paper approach against our constraints quickly Bias for experimentation and building: you would rather ship a working agent pipeline this week and iterate than design a perfect system for a month Independence and ownership. We are a small, fully remote team distributed across Europe (most of us are in EU time zones). You will mostly work async, but you need real overlap with European afternoons for collaboration. Wherever you are based, you need to be available during that window. We welcome applicants from every country. We are building a diverse team with people from many backgrounds While this is a remote position, occasional in-person team meetings may be held a few times per year. Your willingness to travel for these gatherings, as required, is appreciated Who This Role Is Not For Engineers whose agent experience is limited to using AI coding assistants. This role requires building agentic systems, not just working alongside them. People who prefer large, specialized teams with narrow scope. Here you will own problems end to end, from data generation to the performance of the final model.