Generative AI Engineer / GenAI Developer
⚲ Warszawa, Wola
Do uzgodnienia
Wymagania
- PyTorch
- Diffusers
- Transformers
- NumPy
- OpenCV
- BLIP
- CLIP
- LoRA
- DreamBooth
- PEFT
- TypeScript
- Deno
- Node.js
- PostgreSQL
- Kubernetes
- Docker
- React
- ComfyUI
Opis stanowiska
Nasze wymagania:
Production experience shipping generative AI solutions — integrating hosted models (Replicate, OpenAI, Google AI, or similar).
Knowledge of the Flux family and/or SDXL, ControlNet, inpainting, IP-Adapter.
ComfyUI or a comparable node-graph tool: designing your own graphs, debugging pipelines, integrating custom nodes.
Working with LLM APIs: orchestrating Gemini and/or GPT, structured outputs / tool calling, prompt templating, cost and latency control.
Applied Computer Vision: masks, segmentation, layer composition, image preprocessing (resize, format, channels).
Python at ML engineer level: fluency in PyTorch / Diffusers / Transformers / NumPy / OpenCV ecosystem — writing training and evaluation scripts as well as image processing tools.
Training data preparation: curation and cleaning of image datasets, automated tagging/captioning (BLIP, CLIP), quality control, class balancing, augmentations, dataset versioning.
Fine-tuning of generative models: hands-on experience with LoRA / DreamBooth / PEFT for Flux or SDXL.
Mile widziane:
Backend in TypeScript / Deno (Supabase Edge Functions) or Node.js.
PostgreSQL + Supabase (Row Level Security, Storage, Realtime).
Building custom nodes for ComfyUI.
Self-hosting generative models on GPU (Replicate Cog, Modal).
Experience with Kubernetes and Docker for deploying GPU workloads.
React + TypeScript from the API-consumer side.
O projekcie:
You will join an interdisciplinary team building an innovative next-generation digital platform.
Your work will focus on designing, integrating and optimizing generative graphics pipelines based on hosted diffusion models and multimodal LLMs.
Zakres obowiązków:
Designing, optimizing and scaling multi-stage workflows in node-graph GUI environments (ComfyUI) and custom backend pipelines.
Controlling diffusion processes to reduce generation time while preserving visualization quality.
Working with the Flux family (1.1 Pro, Schnell, Fill Pro, Dev Inpainting + IP-Adapter), SDXL and Gemini Flash Image.
Developing automatic mask generation pipeline with model escalation, geometry validation and safe fallback for low-quality results.
Orchestrating Gemini and GPT models for task planning, intent classification, prompt decomposition and input/output sanitization.
Building comparison infrastructure for model variants and parameters, collecting user feedback, automated result analysis and iterative tuning of default configurations.
Oferujemy:
State-of-the-art: work with the latest generative models (Flux, Gemini 3, SDXL) deployed directly to production.
Data foundation: a large, structured internal collection of reference images for pipeline tuning and quality evaluation.
Interdisciplinary team: collaboration with ML engineers, full-stack developers and domain experts.
Modern production stack: real impact on the architecture of a scalable system handling massive real-time graphics requests (Supabase, Kubernetes, Docker).
Production experience shipping generative AI solutions — integrating hosted models (Replicate, OpenAI, Google AI, or similar).
Knowledge of the Flux family and/or SDXL, ControlNet, inpainting, IP-Adapter.
ComfyUI or a comparable node-graph tool: designing your own graphs, debugging pipelines, integrating custom nodes.
Working with LLM APIs: orchestrating Gemini and/or GPT, structured outputs / tool calling, prompt templating, cost and latency control.
Applied Computer Vision: masks, segmentation, layer composition, image preprocessing (resize, format, channels).
Python at ML engineer level: fluency in PyTorch / Diffusers / Transformers / NumPy / OpenCV ecosystem — writing training and evaluation scripts as well as image processing tools.
Training data preparation: curation and cleaning of image datasets, automated tagging/captioning (BLIP, CLIP), quality control, class balancing, augmentations, dataset versioning.
Fine-tuning of generative models: hands-on experience with LoRA / DreamBooth / PEFT for Flux or SDXL.
Mile widziane:
Backend in TypeScript / Deno (Supabase Edge Functions) or Node.js.
PostgreSQL + Supabase (Row Level Security, Storage, Realtime).
Building custom nodes for ComfyUI.
Self-hosting generative models on GPU (Replicate Cog, Modal).
Experience with Kubernetes and Docker for deploying GPU workloads.
React + TypeScript from the API-consumer side.
O projekcie:
You will join an interdisciplinary team building an innovative next-generation digital platform.
Your work will focus on designing, integrating and optimizing generative graphics pipelines based on hosted diffusion models and multimodal LLMs.
Zakres obowiązków:
Designing, optimizing and scaling multi-stage workflows in node-graph GUI environments (ComfyUI) and custom backend pipelines.
Controlling diffusion processes to reduce generation time while preserving visualization quality.
Working with the Flux family (1.1 Pro, Schnell, Fill Pro, Dev Inpainting + IP-Adapter), SDXL and Gemini Flash Image.
Developing automatic mask generation pipeline with model escalation, geometry validation and safe fallback for low-quality results.
Orchestrating Gemini and GPT models for task planning, intent classification, prompt decomposition and input/output sanitization.
Building comparison infrastructure for model variants and parameters, collecting user feedback, automated result analysis and iterative tuning of default configurations.
Oferujemy:
State-of-the-art: work with the latest generative models (Flux, Gemini 3, SDXL) deployed directly to production.
Data foundation: a large, structured internal collection of reference images for pipeline tuning and quality evaluation.
Interdisciplinary team: collaboration with ML engineers, full-stack developers and domain experts.
Modern production stack: real impact on the architecture of a scalable system handling massive real-time graphics requests (Supabase, Kubernetes, Docker).
🔍 Dekoder Ogłoszenia
🟡
Production experience shipping generative AI solutions
Oznacza to, że szukają kogoś, kto faktycznie wdrażał i utrzymywał działające rozwiązania oparte na AI generatywnej, a nie tylko eksperymentował.
🟡
Python at ML engineer level
Wymagają biegłości w Pythonie z naciskiem na biblioteki związane z uczeniem maszynowym, co sugeruje potrzebę tworzenia i modyfikowania kodu związanego z modelami.
🟡
Fine-tuning of generative models: hands-on experience with LoRA / DreamBooth / PEFT for Flux or SDXL.
Oczekują praktycznej umiejętności dostrajania istniejących modeli generatywnych do specyficznych potrzeb projektu, a nie tylko ich używania.
🔴
Self-hosting generative models on GPU (Replicate Cog, Modal).
Sugeruje to, że mogą potrzebować kogoś, kto potrafi zarządzać infrastrukturą do uruchamiania modeli AI na własnych serwerach lub w chmurze, co może być bardziej złożone niż korzystanie z gotowych usług.
🟡
Building custom nodes for ComfyUI.
Oznacza to, że mogą potrzebować kogoś, kto potrafi tworzyć nowe funkcjonalności dla narzędzia ComfyUI, co wymaga głębszego zrozumienia jego architektury.