Machine Learning Engineer – Cloud & Edge AI
Our client builds software for frontline workers operating in highly variable connectivity environments. They are building an architecture that relies on cloud-hosted LLMs for online inference and local SLMs for strictly offline execution across iOS, Android, and Windows. You will engineer the pipelines to make this a reality.
Pozícia už bola úspešne obsadená
Vyberte si z aktuálne dostupných pracovných ponúk, alebo sa zaregistrujte.
Pracovné ponuky
Form of cooperation
TPP
Deployment
Hybrid
Location
Bratislava (SK)
Salary
from 2500
Job description
- Deploy and maintain LLMs on Azure and other cloud infrastructures using inference servers like vLLM or TGI to serve connected mobile clients.
- Build the initial Retrieval-Augmented Generation (RAG) pipelines in the cloud, integrating with Resco’s backend data synchronization systems.
- Transition cloud-proven capabilities to the edge by fine-tuning and quantizing open-weight models (like Llama 3, Phi, or Qwen) for local mobile execution.
- Implement routing logic that switches between cloud AI endpoints and local SLM inference based on network availability and task complexity.
- Benchmark performance, latency, and hardware utilization across both cloud GPU instances and constrained mobile chips.
Prerequisites and skills
- 3 years of software engineering or machine learning experience.
- Experience deploying LLMs to cloud environments. Familiarity with Azure is a strong plus. You should understand model hosting and API routing.
- A working understanding of model quantization (GGUF, AWQ) and the mechanics of shrinking models for local execution.
- Strong proficiency in Python for ML pipelines, plus familiarity with containerization (Docker). Exposure to mobile development, particularly within the .NET ecosystem (C#), is a significant advantage for the edge integration phase.
- An engineering mindset focused on system architecture. You need to und
You will own the lifecycle of AI features, starting from high-capacity cloud deployments down to heavily constrained edge devices. This role requires solving concrete architectural problems regarding offline synchronization, state management, and memory limits across fundamentally different hardware profiles.