About the role
Company
Orcrist builds the Orcrist Intelligence Platform (OIP), a secure, Kubernetes-native data intelligence system deployed as SaaS or self-hosted/on-prem (including air-gapped missions). We fuse data processing, ML, and intuitive UX for defense, law-enforcement, and enterprise teams.
Role
Productionize the NLP/audio/document models that power OIP’s insight experiences. You’ll own model packaging, deployment, monitoring, and evaluation—partnering with Research and product squads to deliver trustworthy enrichment worldwide.
What you’ll do
* Package and deploy models (ASR, translation, OCR, NER, summarization) using Triton/KServe on Kubernetes. * Build evaluation pipelines (WER, BLEU, F1, latency, cost) and automate release gating. * Operate streaming + batch inference via Kafka, Temporal, and backfill tooling. * Monitor drift/quality with Prometheus, Grafana, Evidently; optimize inference cost and performance. * Collaborate with TypeScript teams on payload schemas, contracts, and human-in-the-loop feedback loops.
About you
* 4–8+ years ML engineering/MLOps, shipping models to production. * Strong Python, PyTorch/Transformers, and experience with Triton/KServe or similar. * Comfortable with Kubernetes, GitOps, CI/CD, and GPU workload operations. * Knowledge of evaluation metrics, monitoring, and annotation workflows. * Eligible to work in Germany; export-control screening required for certain programs.
Nice-to-haves
* Temporal, Beam/Flink, or Ray Serve experience; ONNX/TensorRT optimization. * German language (B1+) and familiarity with defense or public safety datasets. * WhisperX, DeepStream/GStreamer, or vector search integrations.
What we offer
* Modern MLOps stack: Triton, Temporal, Kafka, MLflow/Weights & Biases, Evidently, Kubernetes. * Remote-first in Germany with regular Berlin meetups, 30 days vacation, equipment & learning budget.
Originally posted on Himalayas