Data Scientist (1 Year contract)

What matters most

• Contextual engineering AI

  The role develops LLM, generative AI and agentic AI applications for chemical and process industries, including RAG, knowledge graphs and semantic search that turn plant records and engineering knowledge into contextual insights.

• Industrial data foundations

  The posting calls for pipelines that acquire, clean, enrich and contextualize structured and unstructured plant data, maintenance records and operational documents.

• Operational analytics

  The role applies machine learning, deep learning and multimodal AI to predictive maintenance, anomaly detection, process optimization and asset performance.

• Connected workflow automation

  Intelligent agents are expected to interact with enterprise systems, industrial databases, digital twins and engineering tools; APIs and scripting are also used to improve engineering and business processes.

• Production delivery with stakeholders

  Solutions must be scalable, reliable and secure, with deployment, monitoring and maintenance. The role works with global researchers, software engineers, process engineers and business stakeholders from concept through production.

Likely challenges to think through

• Grounding answers in inconsistent industrial sources

  Preparation hypothesis: combining plant data, maintenance records and operational documents may make retrieval coverage and source accuracy difficult to balance. Prepare to explain how you would contextualize records, evaluate retrieved evidence and handle unsupported answers.

• Balancing agent autonomy with operational control

  Preparation hypothesis: agents that interact with industrial tools could save effort but create risks if they act on incorrect context. Prepare a design approach that distinguishes read-only support from actions, with permissions, human review and traceability; the posting names cybersecurity and human-in-the-loop decision support but does not prescribe controls.

• Moving from promising models to dependable use

  Preparation hypothesis: a model that performs well in development may be harder to maintain as plant conditions or data change. Prepare to discuss evaluation with process engineers, production monitoring and when to revise a model rather than automate a decision.

Topics to prepare

• RAG and semantic search for engineering knowledge

  The role develops RAG and semantic search for contextual engineering insights. Revision exercise: sketch how a question about an equipment fault would retrieve relevant maintenance records and operational documents, then explain how you would check whether the answer is supported by those sources.

• Knowledge graphs and industrial context

  The posting asks for experience with knowledge graphs, semantic search, industrial digital twins, or engineering knowledge management systems. Revision exercise: draw a small graph connecting an asset, its measurements, maintenance events and documents; explain what context each relationship adds.

• Agentic AI and system interactions

  The role includes agents that interact with enterprise systems, industrial databases, digital twins and engineering tools. Revision exercise: trace an agent workflow from a user request through a data lookup to a proposed action, identifying where access checks and human review would matter.

• Industrial machine-learning use cases

  The posting names predictive maintenance, anomaly detection, process optimization and asset performance, alongside machine learning, deep learning and multimodal AI. Revision exercise: compare two approaches to an equipment-anomaly problem, including the data needed, how you would evaluate results and the operational cost of false alarms.

• Production reliability and responsible AI

  The role covers secure deployment, monitoring and maintenance; the requirements also name AI governance, cybersecurity and human-in-the-loop decision support. Revision exercise: review a proposed plant-operations AI assistant and list failure modes, monitoring signals and decisions that should remain with an engineer.

Skills to revise

• Prepare and contextualize industrial data in Python

  The posting calls for pipelines covering structured and unstructured data and expertise in Python, pandas and NumPy. As a practice exercise, clean a small illustrative set of plant readings and maintenance notes, align records by asset and time, and document missing or conflicting fields.

• Model and visualize an asset-performance problem

  The role applies analytics to maintenance and anomaly detection and names scikit-learn, Matplotlib and seaborn. As a practice exercise, fit a simple model to illustrative asset data, plot errors or alerts, and explain whether its evaluation would be useful for operations.

• Retrieve data with SQL

  The requirements mention SQL and familiarity with databases such as PostgreSQL or MinIO for storage and retrieval. As a practice exercise, write SQL queries that join asset and maintenance records and return the history needed to investigate one alert.

• Implement a small grounded-retrieval workflow

  The role develops and deploys RAG and semantic search applications. As a practice exercise, use illustrative engineering documents to build a retrieval-and-answer prototype; test questions with relevant, irrelevant and missing evidence, and inspect the cited passages.

• Test an API-backed agent interaction

  The posting requires API proficiency for testing and deployment and describes agents interacting with industrial systems. As a practice exercise, expose an illustrative data lookup through an API using one named example framework, such as FastAPI, then test valid requests, unavailable data and unauthorized requests.

• Plan deployment checks and monitoring

  The role explicitly covers scalable, reliable and secure production AI, including monitoring and continuous improvement. As a practice exercise, write a short test and monitoring plan for the retrieval prototype covering answer support, access failures, changing source data and escalation to a human reviewer.

• Map evidence to the requirements

  Prepare a one-page index of relevant education, experience and projects, with pointers to evidence for Python, industrial AI use cases, knowledge graphs or semantic search, APIs and SQL. Mark gaps rather than inventing examples.

• Rehearse a project walkthrough

  Choose a real project, internship, hackathon or personal project. Prepare a concise account of the problem, data, your contribution, evaluation, results and limitations; explain any relevance to plant operations or asset performance without claiming industrial experience you do not have.

• Sketch an engineering-knowledge application

  As a proposed design exercise, draw a RAG system that uses operational documents or maintenance records for contextual answers. Show ingestion, retrieval, how semantic search or a knowledge graph could help, and how you would check answer quality and source grounding.

• Design a bounded agent workflow

  Prepare a diagram for an agent interacting with an industrial database, digital twin or engineering tool through APIs. Identify permitted actions, validation, human approval points and failure handling; present it as a proposal, not a Yokogawa system.

• Work through an industrial analytics example

  Prepare a small worked example for predictive maintenance, anomaly detection or process optimization. Show how structured and unstructured data would be acquired, cleaned and contextualized, then describe a baseline, evaluation approach and operational limitations.

• Assemble technical implementation evidence

  Select code or notes you can discuss for Python data processing, visualization and model development, plus API testing or deployment and SQL-backed storage. Be ready to explain your actual use of named libraries or frameworks rather than claiming familiarity from the posting alone.

• Explain a path to responsible production use

  Outline how you would deploy, monitor, maintain and improve one proposed solution, including reliability, security, responsible-AI checks and human-in-the-loop decisions. Prepare an example of how you would work with process engineers and other stakeholders from evaluation through deployment.

• Prepare questions for Yokogawa

  Ask which plant-data and documentation sources are in scope, how engineering insights and agent actions would be evaluated, what deployment and governance constraints apply, and what outcomes matter during the one-year contract.