talk-data.com

​LLMs are powerful, but they still hallucinate facts, especially when asked about entities, relationships, or claims that require up-to-date or structured knowledge.

​In this hands-on workshop, we'll explore how to use Wikidata as a grounding and fact-checking layer for LLMs to reduce hallucinations and make AI systems more reliable.

​We'll start with a short introduction to Wikidata and then set up the Wikidata MCP so an LLM can retrieve and verify facts rather than relying solely on its internal memory. This already provides a practical way to ground LLM outputs in verifiable data.

​From there, we’ll go beyond LLM-only approaches and build a small experimental fact-checking pipeline. The system combines semantic retrieval, LLM-based reranking, and natural language inference (NLI) to validate claims against evidence in a more controlled and interpretable way.

​This workshop focuses on evidence-driven verification pipelines that make LLM's reasoning steps explicit and easier to inspect, debug, and improve.

​What we'll cover:

• ​Wikidata as a structured source for factual verification
• ​Setting up and querying Wikidata using MCP
• ​Verifying claims with MCP + an LLM
• ​Moving beyond pure GenAI to evidence-based fact-checking
• ​Finding relevant Wikidata statements with semantic search
• ​Ranking candidate evidence with an LLM
• ​Verifying claims using an NLI model

​​What you'll leave with By the end of the workshop, you'll be able to:

• Ground LLM outputs in structured data to reduce hallucinations
• ​Understand when LLM-only fact-checking is not enough
• ​Build a small, transparent fact-checking pipeline you can adapt to real projects

​About the speaker:

​Philippe Saadé is the AI/ML project manager at Wikimedia Deutschland. His current work focuses on making Wikidata accessible to AI application with projects like the Wikidata vector database and the Wikidata Model Context Protocol.

Join our Slack: https://datatalks.club/slack.html ​

​This event is sponsored by Wikimedia

This two-part discussion series will explore how to make humanitarian spreadsheets more “AI-ready,” bringing together UN OCHA’s new guidance project with real-world lessons from recent AI spreadsheet extraction experiments.

UN OCHA is developing a short, practical guide to help humanitarian teams publish “AI-ready” public datasets that work better with tools like ChatGPT, Copilot, Gemini and open source models like Kimi K2 and GPT OSS running on providers like Groq when users simply upload a CSV or Excel file and start asking questions. The focus is on non-technical users who will not configure agents, write code, or reverse-engineer cryptic column names, but instead expect the AI to correctly interpret the file structure and labels out of the box. By recommending clear naming, consistent tabular layouts, and lightweight documentation, the guidance aims to reduce misinterpretation, hallucinations, and broken analyses when consumer AI tools encounter real-world humanitarian data.

Jan Zheng, a Developer Relations Engineer at Groq who helps people design and build AI prototypes, is exploring exactly these challenges from the model and tooling side. His recent experiments with spreadsheet extraction show that messy, multi-table spreadsheets routinely confuse even advanced models and agent frameworks, leading to unreliable extraction, off‑by‑one errors, looping agents, and high costs. These problems are amplified when complex datasets or vast amounts data are processed by non-technical users of commercial AI tools and open models. Lessons learned through research and usage can inform UN OCHA guidance by clarifying which spreadsheet patterns break current AI tools, which structures make extraction more robust, and how to balance “ideal” AI-ready formats with the messy realities of operational humanitarian spreadsheets.

Over two separate meetup discussions, staff from UN OCHA will introduce the AI‑ready data project in more detail, walk through the specific use case they are targeting, and answer questions from participants about scope, constraints, and potential applications in humanitarian settings. These sessions are designed to surface real-world experiences from practitioners who publish, manage, or use open humanitarian data, and to gather concrete examples of what works and what breaks when datasets are run through consumer AI tools and open source tools running through providers like Groq.

On a following date, Jan will join a dedicated session to react to the project, share his experimental findings on spreadsheet extraction, and discuss how infrastructure choices such as model selection, speed, and prompting strategies interact with the way humanitarian data is structured and published. His perspective will help bridge the gap between guidance aimed at data publishers and the realities of building and tuning AI systems that can reliably interpret messy, real-world spreadsheets used across the humanitarian sector.

This two-part discussion series will explore how to make humanitarian spreadsheets more “AI-ready,” bringing together UN OCHA’s new guidance project with real-world lessons from recent AI spreadsheet extraction experiments.

UN OCHA is developing a short, practical guide to help humanitarian teams publish “AI-ready” public datasets that work better with tools like ChatGPT, Copilot, Gemini and open source models like Kimi K2 and GPT OSS running on providers like Groq when users simply upload a CSV or Excel file and start asking questions. The focus is on non-technical users who will not configure agents, write code, or reverse-engineer cryptic column names, but instead expect the AI to correctly interpret the file structure and labels out of the box. By recommending clear naming, consistent tabular layouts, and lightweight documentation, the guidance aims to reduce misinterpretation, hallucinations, and broken analyses when consumer AI tools encounter real-world humanitarian data.

Jan Zheng, a Developer Relations Engineer at Groq who helps people design and build AI prototypes, is exploring exactly these challenges from the model and tooling side. His recent experiments with spreadsheet extraction show that messy, multi-table spreadsheets routinely confuse even advanced models and agent frameworks, leading to unreliable extraction, off‑by‑one errors, looping agents, and high costs. These problems are amplified when complex datasets or vast amounts data are processed by non-technical users of commercial AI tools and open models. Lessons learned through research and usage can inform UN OCHA guidance by clarifying which spreadsheet patterns break current AI tools, which structures make extraction more robust, and how to balance “ideal” AI-ready formats with the messy realities of operational humanitarian spreadsheets.

Over two separate meetup discussions, staff from UN OCHA will introduce the AI‑ready data project in more detail, walk through the specific use case they are targeting, and answer questions from participants about scope, constraints, and potential applications in humanitarian settings. These sessions are designed to surface real-world experiences from practitioners who publish, manage, or use open humanitarian data, and to gather concrete examples of what works and what breaks when datasets are run through consumer AI tools and open source tools running through providers like Groq.

On a following date, Jan will join a dedicated session to react to the project, share his experimental findings on spreadsheet extraction, and discuss how infrastructure choices such as model selection, speed, and prompting strategies interact with the way humanitarian data is structured and published. His perspective will help bridge the gap between guidance aimed at data publishers and the realities of building and tuning AI systems that can reliably interpret messy, real-world spreadsheets used across the humanitarian sector.

This two-part discussion series will explore how to make humanitarian spreadsheets more “AI-ready,” bringing together UN OCHA’s new guidance project with real-world lessons from recent AI spreadsheet extraction experiments.

UN OCHA is developing a short, practical guide to help humanitarian teams publish “AI-ready” public datasets that work better with tools like ChatGPT, Copilot, Gemini and open source models like Kimi K2 and GPT OSS running on providers like Groq when users simply upload a CSV or Excel file and start asking questions. The focus is on non-technical users who will not configure agents, write code, or reverse-engineer cryptic column names, but instead expect the AI to correctly interpret the file structure and labels out of the box. By recommending clear naming, consistent tabular layouts, and lightweight documentation, the guidance aims to reduce misinterpretation, hallucinations, and broken analyses when consumer AI tools encounter real-world humanitarian data.

Jan Zheng, a Developer Relations Engineer at Groq who helps people design and build AI prototypes, is exploring exactly these challenges from the model and tooling side. His recent experiments with spreadsheet extraction show that messy, multi-table spreadsheets routinely confuse even advanced models and agent frameworks, leading to unreliable extraction, off‑by‑one errors, looping agents, and high costs. These problems are amplified when complex datasets or vast amounts data are processed by non-technical users of commercial AI tools and open models. Lessons learned through research and usage can inform UN OCHA guidance by clarifying which spreadsheet patterns break current AI tools, which structures make extraction more robust, and how to balance “ideal” AI-ready formats with the messy realities of operational humanitarian spreadsheets.

Over two separate meetup discussions, staff from UN OCHA will introduce the AI‑ready data project in more detail, walk through the specific use case they are targeting, and answer questions from participants about scope, constraints, and potential applications in humanitarian settings. These sessions are designed to surface real-world experiences from practitioners who publish, manage, or use open humanitarian data, and to gather concrete examples of what works and what breaks when datasets are run through consumer AI tools and open source tools running through providers like Groq.

On a following date, Jan will join a dedicated session to react to the project, share his experimental findings on spreadsheet extraction, and discuss how infrastructure choices such as model selection, speed, and prompting strategies interact with the way humanitarian data is structured and published. His perspective will help bridge the gap between guidance aimed at data publishers and the realities of building and tuning AI systems that can reliably interpret messy, real-world spreadsheets used across the humanitarian sector.

This two-part discussion series will explore how to make humanitarian spreadsheets more “AI-ready,” bringing together UN OCHA’s new guidance project with real-world lessons from recent AI spreadsheet extraction experiments.

UN OCHA is developing a short, practical guide to help humanitarian teams publish “AI-ready” public datasets that work better with tools like ChatGPT, Copilot, Gemini and open source models like Kimi K2 and GPT OSS running on providers like Groq when users simply upload a CSV or Excel file and start asking questions. The focus is on non-technical users who will not configure agents, write code, or reverse-engineer cryptic column names, but instead expect the AI to correctly interpret the file structure and labels out of the box. By recommending clear naming, consistent tabular layouts, and lightweight documentation, the guidance aims to reduce misinterpretation, hallucinations, and broken analyses when consumer AI tools encounter real-world humanitarian data.

Jan Zheng, a Developer Relations Engineer at Groq who helps people design and build AI prototypes, is exploring exactly these challenges from the model and tooling side. His recent experiments with spreadsheet extraction show that messy, multi-table spreadsheets routinely confuse even advanced models and agent frameworks, leading to unreliable extraction, off‑by‑one errors, looping agents, and high costs. These problems are amplified when complex datasets or vast amounts data are processed by non-technical users of commercial AI tools and open models. Lessons learned through research and usage can inform UN OCHA guidance by clarifying which spreadsheet patterns break current AI tools, which structures make extraction more robust, and how to balance “ideal” AI-ready formats with the messy realities of operational humanitarian spreadsheets.

Over two separate meetup discussions, staff from UN OCHA will introduce the AI‑ready data project in more detail, walk through the specific use case they are targeting, and answer questions from participants about scope, constraints, and potential applications in humanitarian settings. These sessions are designed to surface real-world experiences from practitioners who publish, manage, or use open humanitarian data, and to gather concrete examples of what works and what breaks when datasets are run through consumer AI tools and open source tools running through providers like Groq.

On a following date, Jan will join a dedicated session to react to the project, share his experimental findings on spreadsheet extraction, and discuss how infrastructure choices such as model selection, speed, and prompting strategies interact with the way humanitarian data is structured and published. His perspective will help bridge the gap between guidance aimed at data publishers and the realities of building and tuning AI systems that can reliably interpret messy, real-world spreadsheets used across the humanitarian sector.

En este episodio te contamos con cuáles consejos financieros NO estamos de acuerdo… y por qué. 💛💸

Aquí hablamos de por qué anotar cada gasto puede volverse agotador, por qué un fondo de emergencia de “US$1,000 para todo el mundo” no tiene sentido, y muchos más👀 Dale ME GUSTA 👍 a este video y COMENTA 💬 ¿Cuál de estos consejos sientes que ya no aplica para ti?

🔔 Suscríbete al canal y compártelo con tus amigas, tu pareja o tu familia… para que todos entren al 2025 con las finanzas en orden. 💛

“On Air at Microsoft Ignite” will deep dive into key announcements with expert interviews, demonstrations, and real-world applications for all the latest news. ​​

Hour 2 will feature:​ Azure Infrastructure: Jeremy Winter ​ AI Skills Navigator: Kavitha Radhakrishnan​ Varonis: Shawn Hays​ NVIDIA: Andrew Hester​ GitHub Copilot: Martin Woodward​ AMD: Arjun Oberoi & Daniel Kim

Artisans. Let's meet again, in-person event at a new wonderful event location (Wikimedia Deutschland).

## Talks

• Please suggest talks!

Send us a text In this episode of Data Topics, Ben speaks with Kim Smets, VP Data & AI at Telenet, about his journey from early machine learning work to leading enterprise-wide AI transformation at Telenet. Kim shares how he built a central data & AI team, shifted from fragmented reporting to product thinking, and embedded governance that actually works. They discuss the importance of simplicity, storytelling, and sustainable practices in making AI easy, relevant, and famous across the business. From GenAI exploration to real-world deployment, this episode is packed with practical insights on scaling AI with purpose.

A talk about managing anxiety, burnout, and career decisions in tech. Kim Scott discusses emotional agility, thought models, and coaching that helped her align with her values, stay calm, and make bold career moves. The session includes a personal journey through data science and related roles, with practical tools to stay focused and authentic.