Senior Machine Learning Engineer
Pubblicato il 13-08-2026 - Skillvue in Firenze
We are a fast-growing HR tech startup backed by leading international VCs, having raised €9M+ from 360 Capital, IFF, Kfund, and 14Peaks. We are a team of 30+ professionals passionate about shaping the future of talent assessment. Skillvue is a Skill AI Assessment platform (SaaS) to hire top-skilled candidates and measure employee skill, culture, and leadership at scale to upskill and grow the workforce by leveraging AI. Our platform enables companies to conduct Skill AI Assessments for both external and internal hiring, as well as targeted evaluations across their entire workforce. Role overview You will own end-to-end ML systems: model training, fine‐tuning, deployment, monitoring, and cost/performance optimization. Partner closely with organizational psychologists, people scientists, and software engineering to productionize LLMs, real‐time conversational agents, and ML pipelines. You will report to the Head of AI & Science and drive engineering best practices, reliability, and reproducibility across the stack.
Design, build, and maintain end-to-end ML platforms and pipelines:
data ingestion, feature engineering, training, validation, deployment, and monitoring.
Develop, fine‐tune, and deploy LLMs and GenAI services for assessment tasks (prompt engineering, instruction tuning, RLHF/IL, retrieval‐augmented generation).
Implement scalable, low‐latency inference systems (serverless and/or containerized), real‐time voice/text conversational agents, and batching strategies for cost‐effective throughput.
Build infrastructure‐as‐code (Terraform/CloudFormation) for reproducible environments and secure, compliant deployments.
Create automated CI/CD for data, models, and infra (model/data versioning, reproducible training runs, canary/blue‐green deployments).
Optimize model size and inference cost using quantization, pruning, distillation, sharding, and hardware‐aware optimizations.
Implement monitoring, observability, drift detection, and alerting for model performance and