• Design and develop performance testing frameworks for deep learning compilers and workloads • Build and maintain automated pipelines (CI/CD) to continuously track performance across models, hardware, and compiler changes • Implement benchmarking systems to measure latency, throughput, and efficiency of AI and HPC workloads • Analyze performance trends over time and identify regressions, bottlenecks, and optimization opportunities • Partner with compiler and architecture teams to debug and resolve performance issues • Develop tools and dashboards for performance visualization, reporting, and insights • Enable scalable testing across diverse GPU systems and environments • Improve infrastructure to ensure reliable, reproducible, and high-signal performance data

Role Description Incubate and validate new ML initiatives end-to-end. On Innovation, you’ll build adoption-ready prototype vertical slices spanning data flows, model serving, evaluation, and product integration—then hand off clear artifacts so delivery teams can productize and own them long-term. - Build ML prototype vertical slices that connect ingest/processing to inference and visible product outcomes (search, insights, UX flows). - Create evaluation harnesses and decision artifacts: datasets, baselines, quality/latency/cost metrics, and go/no-go recommendations. - Package prototypes for adoption: containerize services, define reproducible deployments, and produce runbooks/checklists. - Partner with Research and Data Engineering on dataset curation, annotation loops, experiment tracking, and safe iteration. - Make prototypes operationally credible: instrumentation, monitoring, and security/compliance basics (PII handling, provenance mindset). Qualifications - 3+ years ML engineering/MLOps experience (level dependent), with evidence of shipping real systems. - Strong Python and hands-on PyTorch/Transformers; comfortable taking models from notebook to reproducible services. - Practical Kubernetes + containers experience; able to deploy and troubleshoot in production-like clusters (including offline/air-gapped constraints). - Strong evaluation discipline and monitoring mindset; comfortable communicating tradeoffs clearly. - Eligible to work in Germany; EU/NATO citizenship preferred and export-control screening applies. Nice‑to‑haves - GPU serving/optimization experience (Triton/KServe, ONNX/TensorRT, batching, quantization). - Streaming/pipeline tooling (Kafka, Ray, Beam/Flink/Spark) and search/vector/graph integrations. - German language (B1+) and/or experience with regulated/public-sector datasets and workflows. Benefits - Modern ML stack in real constraints: Kubernetes, streaming, and hybrid/on-prem/air-gapped deployments. - Remote-first in Germany with regular Berlin workshops, 30 days vacation, equipment & learning budget. - High leverage: your prototypes and handoffs unblock multiple delivery teams.

• Conduct structured process audits to document workflows, identify inefficiencies, and quantify manual effort costs. • Develop and prioritize a roadmap for process improvements and tool integrations based on impact assessments. • End-to-end design and implementation of solutions, including but not limited to AI agents, API integrations, and custom internal tools. • Seamlessly integrate existing tools like Slack, Jira, and Google Workspace to enhance system interoperability. • Define success metrics, monitor the adoption of new processes, and iteratively improve solutions based on feedback. • Contribute to and expand the organization’s process automation playbook for future initiatives.

• Work directly with clients to identify their highest-impact problems • Translate client problems into structured approaches • Design and deploy scalable, AI-powered solutions • Operate side-by-side with clients building in real time • Iterate quickly and demonstrate value throughout the engagement • Define, solve, and ensure solutions are adopted in high-stakes environments