自由数据科学家（本体论和图专业知识）
Freelance Data Scientist with Ontology and Graph expertise
adesso Belgium 正在寻找一名数据科学家（2名），以支持我们的客户构建 AI 代理平台的数据基础。
客户信息：
我们的客户是国际性的独立间接采购服务提供商。
要求
期望开始日期：1名 ASAP，1名 2027 年 1 月
时长：12 个月
薪资：根据您的所在地而定
时区：CET
您的职责
构建支撑客户采购 AI 代理平台的数据和上下文层——标准模型、知识图谱、语义/分类层，以及代理依赖的检索和评估流程。从第一天起就是生产级基础设施，而非概念验证：版本化、测试并通过多租户设计。
您将参与以下工作：
- 数据产品与标准建模 —— 从源数据摄入到标准模型，端到端地设计并交付首个参考数据产品（供应商主数据、支出立方体、合同登记）。
- 知识图谱与本体 —— 搭建并运营平台和工作图谱（本体、参考数据、租户隔离）；负责图谱与向量检索设计，根据每个数据源决定数据是以实体/关系、嵌入还是两者结合的方式表示。
- 主数据与语义层 —— 为供应商、合同和类别构建匹配规则、黄金记录和管理流程；参与业务术语表和分类到本体的映射。
- 提取与评估 —— 构建文档提取流程（例如来自 SharePoint 的数据源），带有置信度评分和对不确定情况的人工审核；构建并维护标注的评估数据集，并根据约定的阈值进行准确率基准测试。
- 连接器框架 —— 构建源连接器（从 Fabric 数据产品和主数据黄金记录开始），以便新数据源可以接入而不修改图谱核心或查询服务。
- 质量反馈循环 —— 实现机制，将代理的覆盖、失败或低置信度检索反馈给数据所有者，使数据产品随时间不断改进而非退化。
- 工程实践 —— 仅在 GitHub/Azure 环境中工作；通过 Pull Request 发布，包含 CI（构建、测试、安全扫描、评估固定装置）；对所有模式、连接器契约和提取配置进行版本控制。
所需经验
硬技能：
- 具备将知识图谱系统投入生产的实际经验（不仅仅是原型开发）——包括模式设计、查询性能、租户/数据隔离 - 使用 Neo4J 和 Python
查看英文原文
adesso Belgium is looking for a data scientist (2) to support our customer in building an AI Agent Platform Data Foundations.
Customer info:
Our customer is an international, independent provider of indirect procurement services.
Requirements
Preferred start date: 1xASAP and 1x January 2027
Duration: 12 months
Rate: depending on your location
Time Zone: CET
Your role
Build the data and context layer that powers the customer's procurement AI agent platform — canonical models, knowledge graph, semantic/taxonomy layer, and the retrieval and evaluation pipelines agents depend on. This is production infrastructure from day one, not a proof of concept: versioned, tested, and built for multiple tenants.
You will work on:
- Data products & canonical modeling — Design and deliver the first reference data products (Supplier Master, Spend Cube, Contract Register) end to end, from source ingestion through the canonical model.
- Knowledge graph & ontology — Stand up and operate the platform and working graphs (ontology, reference data, tenant separation); own the graph-plus-vector retrieval design, deciding per source whether data is represented as entities/relationships, embeddings, or both.
- Master data & semantic layer — Build matching rules, golden records, and stewardship processes for Suppliers, Contracts, and Categories; contribute to the business glossary and taxonomy-to-ontology mapping.
- Extraction & evaluation — Build document-extraction pipelines (e.g., from SharePoint sources) with confidence scoring and human review for uncertain cases; build and maintain labelled evaluation datasets and accuracy benchmarks against agreed thresholds.
- Connector framework — Build source connectors (starting with Fabric data products and master data golden records) so new sources can be onboarded without modifying the graph core or query service.
- Quality feedback loop — Implement the mechanism routing agent overrides, failures, or low-confidence retrievals back to data owners so data products improve over time rather than decay.
- Engineering practice — Work exclusively in the GitHub/Azure environment; ship via pull request with CI (build, tests, security scan, evaluation fixtures); version all schemas, connector contracts, and extraction configs.
Required Experience
Hard skills:
- Demonstrated experience taking a knowledge graph system to production (not just prototyping) — schema design, query performance, tenant/data isolation - Using Neo4J and Python
- Strong background in semantic modeling: ontologies, taxonomies, business glossaries, and their relationship to graph and canonical data models.
- Hands-on experience with semantic web standards and technologies including OWL/OWL2, RDF, RDFS, SPARQL, ontology lifecycle management, taxonomy alignment, and semantic reasoning.
- Experience building data products and master-data pipelines (entity matching, golden-record consolidation, stewardship workflows).
- Experience designing and implementing hybrid retrieval architectures combining graph databases (e.g., Neo4j) and vector databases to support semantic search, RAG, and AI agent workloads.
- Experience with retrieval architectures combining graph and vector/embedding approaches.
- Experience building and maintaining labelled evaluation datasets and measuring extraction/model accuracy against a threshold.
- Proficiency with modern data/ML engineering practices: version control, CI/CD, automated testing, containerized deployment
- Experience with Microsoft Fabric and SharePoint-based document extraction pipelines.
Soft skills:
- Fluent in ENG
- Open to travelling if needed (EU), this position is fully remote
- Collaborative mindset, ideally with a proven similar work experience
Why join adesso Belgium?
- Work in an international group with strong local expertise in Belgium.
- Collaborate with cross-functional teams to deliver trusted, compliant, reusable data products.
- Flexible freelance setup with the support of a growing adesso Belgium community.
Originally posted on Himalayas