Data Scientist, ADAS Analytics Machine Learning

Job Description - Data Scientist, ADAS Analytics Machine Learning (MER
00048Z5)

Description

At Mercedes-Benz Research & Development North America (MBRDNA), we are committed to delivering world-class automotive technologies that push the boundaries of what is possible. Our teams of highly skilled engineers and designers use cutting-edge software and technology, to enhance the driving experience and reduce environmental impact.

As Mercedes-Benz scales ADAS across production fleets, the US ADAS Data & Forensics team is building applied-AI capabilities that go beyond descriptive KPI reporting. These capabilities include vision-language models that analyze SSR Scene Safety Recording footage, LLM-based event classification and reasoning, embedding-based semantic retrieval over driving scenarios, automated scenario discovery, natural-language event descriptions, and model-based ranking that surfaces the most important events from thousands of daily drives.

As a Data Scientist, you will develop and deploy these capabilities in production for two stakeholder groups: management, through fleet-performance summaries, trend forecasts, and AI-generated event narratives; and engineers, through granular model-assisted analysis of ADAS calibration, scenarios, edge cases, and behavioral patterns across road types, weather, firmware, and driver cohorts.

Relocation assistance (domestic or international) is not provided for this position.

• Apply vision-language models to SSR video and combine video analysis with structured telemetry to create multimodal event representations.
• Build LLM classification and reasoning pipelines that triage events by severity and root cause and generate human-readable summaries of takeovers, safety events, and deactivations.
• Design embedding pipelines and semantic search for similar-event retrieval, and develop unsupervised clustering methods that discover recurring scenarios and edge-case families at fleet scale.
• Build model-based ranking and scoring systems that reduce manual event triage, develop active-learning loops using engineer feedback, and proactively detect fleet-level anomalies.
• Quantify the impact of firmware updates and configuration changes on KPIs, segment driver cohorts, and design A/B and quasi-experimental frameworks.
• Analyze campaign effectiveness, build coverage-optimization models, and develop automated fleet-quality scoring.
• Deploy models into PySpark and Delta Lake pipelines and the FastAPI analytics API, build evaluation frameworks for foundation-model outputs, and operate on the Azure data platform, including ADLS, Synapse, and Container Apps.

• Bachelor's or Master's degree in Data Science, Machine Learning, Statistics, Computer Science, or a related quantitative field. A Master's degree or PhD is beneficial but not required; demonstrated experience carries equal weight.
• 3+ years of professional experience.
• 2-5 years of experience in data science or applied machine learning.
• Depth in at least one of the following: applied foundation models such as LLMs, VLMs, or embeddings; classical machine learning in production; or statistical experimentation.
• Strong Python skills using Num Py, pandas, and scikit-learn, with an emphasis on clean, testable code, and strong SQL skills.
• Experience with ML model development, including feature engineering, model selection, and evaluation on real data.
• Solid statistics knowledge, including hypothesis testing, regression, and experimental design.
• Ability to communicate findings clearly to engineers and management through reports, presentations, and dashboards.
• Ability to turn ambiguous questions into structured analytical approaches.

• LLM or VLM application experience, including prompt engineering, structured outputs, and evaluation.
• Embedding models and vector similarity for retrieval or clustering.
• PySpark for large-scale processing; candidates with strong pandas experience may ramp up.
• Time-series analysis or anomaly detection.

• Multimodal foundation models applied to video or image data.
• RAG or vector-database systems such as FAISS, pgvector, or Pinecone.
• Spatial or geospatial clustering with DBSCAN or HDBSCAN.
• Ranking or recommendation systems, active learning, PyTorch, or Tensor Flow.
• Delta Lake or Parquet;
  FastAPI or model-serving APIs; MLOps platforms such as MLflow or Weights & Biases.
• Cloud-platform experience in Azure, AWS, or GCP.
• Vehicle telemetry or automotive-domain experience;…