Senior AI Engineer

About the Position

Join Coinis as a Senior AI Engineer to lead the generative pipeline for ad creatives, leveraging diffusion and transformer-based models.

Responsibilities

• Own the generative pipeline end to end. Image, video, and multimodal.
• Build a model router that picks the best model per task behind the scenes.
• Design prompt templates that encode brand voice and visual style from our Brand Profile system.
• Ship post-processing that turns raw model output into production-ready ads.
• Partner with product to close the gap with AdCreative.ai and beyond.
• Run evals. Measure win rate against human designers and competitors.
• Keep cost per generation in line with token economics.

Requirements

• 5+ years in ML or applied AI. Production experience shipping generative systems.
• Deep experience with diffusion, transformer-based image and video models.
• Strong Python. Comfortable in PyTorch or JAX.
• Experience running model inference at scale on GPUs.
• Prompt engineering for real products, not demos.
• Taste. You know when output looks generic and can fix it.

Nice to Have

• Fine-tuning or LoRA experience on creative models.
• ControlNet, IP-Adapter, or similar conditioning workflows.
• Experience with Cloudflare Workers, R2, or edge inference.
• Ad creative or marketing domain background.

Qualifications

Education

BSc or MSc in Computer Science, ML, or related. Equivalent experience also welcome.

Experience

5+ years

Skills

• Python
• PyTorch
• Diffusion Models
• LLMs
• Prompt Engineering
• GPU Inference
• Cloudflare Workers
• ComfyUI
• Stable Diffusion ecosystem
• Evals

We Offer

• Competitive salary in EUR. Benchmarked against EU senior IC bands.
• Equity participation in the SaaS spinout.
• Direct ownership of the core product differentiator.
• A founder and leadership team that ships.
• Flexible remote work. EU hours.
• Hardware budget. Conference budget. Learning budget.

About the Company

Coinis turns a product URL into dozens of ad creatives. You own the model layer that makes that happen. You pick the right image and video model for each job, route requests, tune prompts, and push quality higher every sprint. Our users are ecommerce operators, not prompt engineers. They should never see a slider, a seed, or a negative prompt.