Megan Risdal

Yesterday, I had the pleasure of presenting our work on image captioning evaluation at IJCAI 2025 in Montreal 🇨🇦. In this work, we analyze how evaluation methods for image captioning have evolved over time, and we look ahead to how these metrics can better capture the richness and complexity of captions generated by the new wave of Multimodal Large Language Models (MLLMs). To support the community, we are also releasing a GitHub repository that brings together existing evaluation metrics in one place — with the goal of making it easier for researchers to explore, compare, and advance caption evaluation methods. 📄 Paper: https://lnkd.in/ejqqCMqg 💻 Repository: https://lnkd.in/e5ySn9kf A huge thank to Marcella Cornia and Rita Cucchiara! #IJCAI2025 #ImageCaptioning #Evaluation #MLLM #AIResearch

102

1 Comment

A common question I'm frequently asked: "Which AI agent framework do you use for building AI apps?" Initially, I jumped into LangChain and LlamaIndex, quickly got intrigued by CrewAI, and lately, I'm diving deeper into Agno—each great in their own way. With OpenAI and Google also releasing their own AI agent frameworks, the landscape keeps changing fast. But here's what's interesting—I recognized early on that no matter which framework I selected, an AI agent is only as powerful as the tools it leverages. And keeping tools modular, reusable, and cross-project compatible turned out to be a real engineering headache. Every new AI project felt repetitive—building similar tools again and again was neither efficient nor scalable. Then I came across MCP: the Model Context Protocol, an open-source protocol framework developed by Anthropic. Think of MCP like HTTP—but instead of websites, it makes AI agents and tools universally connectable. Although I've experimented with MCP before, yesterday it got enabled on my go-to automation platform: n8n. I quickly spun up an MCP server on n8n, populated it with custom-built tools, existing utilities, and even embedded n8n workflows themselves as reusable AI tools. Now I have a single, cohesive "toolbox" server that I seamlessly integrate across multiple AI projects—be it Cursor, Claude desktop, or my own custom agents built on LangChain or Agno. If you're building AI-driven products or workflows, I'd highly recommend exploring MCP for tool interoperability. MCP feels like true engineering efficiency—it lets me stop reinventing wheels and finally spend time on real innovations. #GenAI #AIAgents #MCP #n8n

691

55 Comments

Demo: how to surface and use Unity Catalog Metric Views in Databricks AI/BI — to enable self-serve access to trusted business metrics. See how to set up central governance with row-level security and column masking, and more.

403

8 Comments

It would've taken 25 years for this Snowflake customer to get 100% business data quality coverage. With Anomalo, Discover now configures unsupervised data quality monitoring within 10-15 minutes for each dataset. Learn how Discover’s data team flipped the script using Anomalo on Snowflake to scale trust across petabyte-scale data: https://lnkd.in/gPjxDCFH

11

Is our growing reliance on AI changing not just how we write, but how we think? I recently discussed this with CBC News for an article on this critical question. We discussed the growing "homogenization" of language online—a trend my lab at USC, and specifically Zhivar Sourati, has been tracking. In the piece, I touch on our research findings, which show a measurable drop in complexity variance of written text since the widespread adoption of LLMs. We're seeing a feedback loop where humans are beginning to mimic AI's linguistic style, potentially limiting our own creativity and reasoning. The real danger here isn't just dull writing; it's the erosion of the cognitive diversity that defines human society. Our new review article dives deep into this threat (https://lnkd.in/gNq7Rzgx). Read the full article here: https://lnkd.in/gKUnUPdh 

96

4 Comments

Data teams, your world is shifting fast. But how much? To answer that, we just released Hex's first-ever State of Data Teams Report. Think of it as a pulse check on what's top of mind for data teams right now. Here's a peek:   🤖 77% 𝐛𝐞𝐥𝐢𝐞𝐯𝐞 𝐢𝐧 𝐭𝐡𝐞 𝐩𝐨𝐭𝐞𝐧𝐭𝐢𝐚𝐥 𝐨𝐟 𝐀𝐈; 𝐛𝐮𝐭 𝐨𝐧𝐥𝐲 3% 𝐡𝐚𝐯𝐞 𝐢𝐭 𝐚𝐭 𝐭𝐡𝐞 𝐭𝐨𝐩 𝐨𝐟 𝐭𝐡𝐞𝐢𝐫 𝐩𝐫𝐢𝐨𝐫𝐢𝐭𝐲 𝐥𝐢𝐬𝐭. What's behind that gap? 📊 70% 𝐚𝐫𝐞 𝐫𝐞𝐭𝐡𝐢𝐧𝐤𝐢𝐧𝐠 𝐭𝐡𝐞𝐢𝐫 𝐝𝐚𝐭𝐚 𝐬𝐭𝐚𝐜𝐤𝐬. BI dashboards promised self-serve analytics — why haven't they delivered? 📈 90% 𝐨𝐟 𝐝𝐚𝐭𝐚 𝐭𝐞𝐚𝐦𝐬 𝐚𝐫𝐞 𝐡𝐢𝐫𝐢𝐧𝐠 𝐨𝐫 𝐡𝐨𝐥𝐝𝐢𝐧𝐠 𝐬𝐭𝐞𝐚𝐝𝐲. How can data leaders strategically build their teams with AI in mind? We dig into these topics and more in our full report ➡️ https://lnkd.in/gzY8vBNw #dataanalytics #BI #datascience #datateams

55

3 Comments

Analytics Governance isn’t optional in the Public Sector. It’s urgent. From misconfigured dashboards to manual QA bottlenecks, public agencies face mounting pressure to deliver trustworthy data analytics. In a world where data drives decisions and public trust, governance at the Analytics layer is no longer a nice-to-have. 🔍 Discover the 4 key reasons why Analytics Governance must be part of every public agency’s strategy, and how Wiiisdom helps make it happen. 📖 Read the full article at https://lnkd.in/ehpjvPXD #AnalyticsGovernance #governance #decisionmaking #PublicSector

7

Snowflake’s best practices for Agentic AI development (90% of teams will still skip these basics, fail, and blame the tech) Last week I was at Snowflake’s AI World Tour event in Toronto. They went deep on what it takes to move past hype and actually make AI agents work in SMB & enterprise. 📌 Here are the takeaways worth paying attention to: 1) Owners in the loop Every agent needs a business owner (outcomes) and a data owner (quality). Without that partnership, you’re building a toy, an expensive one. 2) Start small Pick one business function, build the agent, test, and improve. Scale only after you see value. 3) Semantic Model first Context is king. Add data to your semantic model before you start chasing bigger ones. **Semantic model = dictionary + rules + context that makes data usable and trustworthy for the agent in your specific scenario 4) Playing with a Model's size Bigger is not necessarily better. Sometimes splitting semantic views makes it easier for your agent to call across your data. 5) Golden Datasets Lock down reliable “truth sources” that are undeniable. Otherwise, as you add functionality, the experience will regress and hallucination will increase. 6) Test properly Run a pilot with a small user set. Agents will fail in ways you don’t expect. Learn early. Very similar framework to what we preach at Tecna, which has translated into clients generating $20,000 to $750,000+ in recurring revenue and savings.. It’s refreshing to hear. TL;DR > Make sure that the people executing AI in your company are paying attention to: clarity, ownership and iterative testing focused on value (if you want to succeed) The ROI is real when you follow the fundamentals. #SFWT2025 #AI #Data #AgenticAI #Snowflake

8