AI @ Alliander Career Alliander Business IT & Management Solution Architect Data & AI @ Consultancy @ Capgemini & Alliander Nextview Gas Optimalisation Tool Holonet Holocube Data & AI Datacloud Voltix AI Previous customers Personal IT Stuff… v oltviz.com linkedin.com/in/ sanderdw github.com/ sanderdw /hassio - addons onabodyjourney.com onaspacejourney.com github.com/ sanderdw
of data and transactions Alliander will make it easier for consumers, businesses and public authorities to access and share relevant data. This will open up opportunities for new services, such as dynamic capacity tariffs. https://www.alliander.com/ - /media/files/corporate/digitalisering/digitalisation Co - creation through collaborations and partnerships Alliander will work together with DSOs, knowledge partners and the business community to prevent double work, benefit from economies of scale and use digitalisation capacity more effectively. - vision - and - strategy - alliander - 2026 - 2030.pdf?v=1& d=20260722T144248Z Artificial intelligence In 2030 AI will be the key to optimal performance in the new energy system. We will use AI as a power tool in the hands of our employees and set clear priorities for how it is used.
for Alliander A DECENTRILIZED WAY OF WORKING WITH DATA 01 02 Domain - Oriented Ownership 03 Self - Service Data Platform Domains (referred to within Alliander as core registries) are functional partitions of the data landscape that contain business critical data. Core registries are not organizational structures. Ownership of data resides within organizational units, ensuring that the people who know the data best are also responsible for it. - Data as a product Data is treated as a product , and the user is treated as a customer . Customers can influence the offering of data products through agreements with the data provider on content and quality. Data products have clear ownership and are reusable across multiple use cases. The self - service data platform is a combination of technologies that technically enable anyone to produce and consume data products. The platform includes all technologies required for ingestion, ETL, storage, and visualization needed to transform data into insights.. OWNERSHIP BELONGS TO THE PEOPLE WHO KNOW THE DATA BEST CONSUMERS DEFINE WHICH DATA PRODUCTS AND QUALITY LEVELS ARE NEEDED TO CREATE BUSINESS VALUE DATA PRODUCTS ARE SELF - DESCRIBING AND DECOUPLED FROM SOURCE SYSTEMS DATA PRODUCTS ARE REUSABLE ACROSS DIFFERENT USE CASES 04 Federated Computational Governance Data governance is guided centrally by the Data Office, for example by defining roles and responsibilities, but is executed in a decentralized manner by organizational units and domains. This ensures a consistent way of working with data while keeping responsibility with the people who work with the data on a daily basis.. 18 / 22
capability’s ALLIANDER VOLTIX AI DATACLOUD DATA PRODUCTS DATA CATALOG EXPLORATION REPORTING MLOPS APPS TRANSFORMATION CONVERSATIONAL ANALYTICS DATA CONSUMERS BUSINESS SOLUTIONS DATA PRODUCERS ENTERPRISE AI DATA SOURCES INGESTION
Publication Layer A fully automated data product publication and consumption architecture that enables producers and consumers to exchange data products in a secure , entirely self - service manner via a central catalog . It is built on a minimal set of selected technologies to support APIs, bulk structured and unstructured data, streaming, and dashboard/reporting interfaces, without relying on, for example, a platform team. P rinciples (simplified…) • Scope: Analytical/data integration (Data Products), not process/transactional integration • Scope: multi - source • No central storage • Automated publishing and consumption via the catalog (authorization) • Minimize the number of technologies • Technology should not be in competition with each other • Cover 90% of the desired producer and consumer functionality • Producer/provider must be able to publish with minimal extra effort • Easy consumability by the consumer • Data product metadata as part of the Data Product (JSON) Metadata (Data Product name, responsible party, BIVP, description) should ideally be included with the Data Product (e.g., think of a JSON metadata endpoint in the API, VARIANT column in Snowflake, or a JSON object in S3). • Pay - per - use: Compute costs should be borne by the consumer where possible • Secure by design and supportive of the Data & Governance policy (decentral ownership) • Integration with the Data Catalog • Not the EPL, but the provider is responsible for specific Data Product governance (versioning, archiving, cleaning, and management of published Data Products).
data products Findable - No ‘or der’ button - Business content in proprietary systems (SAP, Workday, Etc ) - Governance and authorization handled elsewhere - You need to integrate yourself Findable & Accessible VS - EPL as foundation Can have multiple interfaces Accessible by design Governance and authorization are part of the user experience
brainer Automation is all we need name: "Example Product using SDK" user_description: "This is an example product created using the Atlan SDK." data_domain: Digitalisering daap_status: Active daap_visibility: Public certificate_status: Draft certificate_status_message: "This certificate is in draft status." metadata: Owner: ART: Data & Analytics Team: BI & Governance Tooling ServiceNow Support Group: aabbccddeeff00998877665544332211 Data Classification: Confidentiality: C1 Integrity: I1 Availability: A1 Privacy: No - does not contain personal data Business Attributes: Data Product Type: "Consumer Aligned" BIA: true ETC…
, Apache Iceberg Sander de Wildt For years, we’ve relied on REST and “ is doing for data . OpenAPI specs to abstract software so any service can talk to any application. Today, Apache Iceberg is doing the exact same thing for structured data, it provides an open (including the way you want to run it) and standard interface so that no single engine owns your data any mor e. ☺
Enabling ‘on prem’ data products, portability, sovereignty Checking the EPL P ALLIANDER Enterprise Publication Layer rinciples • Suitable for our (highly) secured environments like OT/IT • Producer and consumer can each choose their own technology independently • Key pattern for data exchange outside Alliander • It compliments our Snowflake environment using the Iceberg integration (remember SAP HANA virtual tables anyone?) Also • Open Source (for real…) • One of the answers to portability • Business Critical use cases can leverage bulk tabular interfaces • Minimizing vendor lock - in • We already leverage some Apache Iceberg interfaces
DataCloud Voltix AI Self - service, secure, pay - per - use platform: data products, catalogue, orchestration, reporting. Alliander's AI platforms: MLOps , agents, managed MCP servers and model access. 2 3 4 5 Apache Iceberg API Data Product Extended Data Sharing AI Synthetic Data Unstructured Data Products E2E Integrated into the catalog Catalog integration for data product designed API’s Sharing outside Alliander with DSO’s or other partners As as service (using production as reference data) As data product?
Solution Architects AI Business Support AI Engineering Support AIX Co -development AI use -cases & AI Platform MLOps PAX Microsoft 365 & personal productivity apps AABD Data professionals spread across the organization who support with ML use -cases RCDT Experimental AI techniques and long -term AI research APINT, DevX & Jarvis Developer AI
CLI to train and run ML models VOLTIX MLOPS • There was no central way of working for ML • Currently cloud native and leaning on AWS Sagemaker • Forge CLI Abstraction and can be extended to support on operations - prem / critical
• • • • • Alliander Power of Pydantic AI Open Source Deployable on Kubernetes (PAX) and AWS Bedrock and LLM Proxy integration Observability with Dynatrace API/A2A/MCP Management with Gravitee IAM Act On Behalf Roadmap • Evals • Leverage Unstructered data sources • Agent as a service
Productivity Enterprise (agentic) AI • • • • • • • • • • • • • • Low/No Code Connectors No business - critical processes, or processes that may become business - critical For yourself or your immediate team Maximum of 15 users No sensitive data (BIV classification lower than 2) You remain responsible for your own solution VS Risks of point solutions? UX? Grey zones? To be further investigated… High Code Open Protocols Supports business processes Sensitive data, supported by the existing ecosystem Connectivity within and outside Alliander PAX (Kubernetes) and AWS support Security compliant To technical?
EPL GenAI - ready? Ideally ALLIANDER Enterprise Publication Layer • Ask functional data questions about data products, in any form (EPL query / bulk / streaming / unstructured). • Help data stewards : data - quality agents that use both metadata and data . • Enable applicative / agentic access to the EPL , so it can plug into Alliander processes. Challenges • • • • • • • • • • • General Data and AI literacy The current semantic layer state GenAI architecture knowledge needs time to sink in Platform readiness, especially knowledge platforms Make the Alliander data policy LLM - readable Make base registers & glossaries LLM - readable Write data - quality rules LLM - readable Vendors: not only REST APIs but MCP tools/agents Choosing the right LLM model(s) Evaluation Prompt engineering is essential
looked at three (obvious) scenarios 1 2 3 · short term Purely AI - driven Traditional Reporting Questions are only answered directly via AI and semantically rich data products; classic dashboards become irrelevant fast . Power BI stays the core for analysis and dashboards, with broad adoption by non Need Hybrid (AI + BI) Reporting / dashboarding - IT users. AI conversations wil increase and traditional dashboards wil decrease over a longer period Conversational BI/Analytics Operational monitoring + – Known questions, repetitive answers + – Shared context for meetings + – Graphs / visualisations + Ad - hoc exploration – + Cross - system analysis – + Prediction + – + With the “shift left” of semantics, ad - hoc exploration and cross - system analysis can lean much more on Conversational BI. People explore data without Power BI skills, and iterate on the chart they need themselves.
queries, evaluation built Semantic models sits between the data layer and every AI / BI consumer. Dimensions, synonyms metrics, filters and verified quer y’s - in Evaluation datasets for agents, built from real production questions. Traces : every planning step and MCP call, with timings.
decides how to show the answer, not only what . BACK - END Atlan - Metadata Broad context like domains, business terms, AI Ready data check. FRONT - END AGENT CopilotKit UI Voltix Agent ( Pydantic AI) Snowflake C hat interface. Runtime with openGenerativeUI: true renders agent - generated HTML/CSS/JS (Chart.js ). Slim agent with generative - UI instructions, queries Atlan for broad context and Snowflake for specific data questions via MCP. Leveraging build - in Text - to- SQL via the semantic models, dimensions, metrics, verified queries, custom instructions and some evaluations. – Structured Data - Governance & Agent registration - Evaluation Framework Dynatrac e – Agent Obervability OTEL, End - to- end agent trace, LLM call latency, Tool call details, Token usage, Error visibility. Gravitee – Agent Ingegration Agentic enterprise AI integration, usage plans, LLM Proxy, MCP guardrails.
SQL came from the curated set in the semantic model, not from free generation. What the user sees • A reliability badge: SQL verified query or generated
Voltix AI Self - service, secure, pay - per - use platform: data products, catalogue, orchestration, reporting. Alliander's AI platforms: MLOps , agents, managed MCP servers and model access. 2 3 4 5 Voltix Forge ( MLOps ) Conversational BI Voltix Vibe Extended Vision Models 3D Vision Realtime Inference From MvP to Production and iteration Enterprise governed citizen development Segment Anything, individual object clustering Point Clouds
mark3t1ng
jonoalderson
csswizardry
szymonslowik
aleyda
schroneko
matthewcrist
codingconduct
apolaine
charlesmeaden
tammyeverts
chriscoyier
tmiket
