Alessandro Mariani

What Anthropic, Cohere, and Databricks know: RAG beats fine-tuning at scale… Fine-tuning is old money. RAG is smart money. Teams keep burning compute retraining models. Then act shocked when accuracy still drifts. Fine-tuning locks knowledge in stone. Your data moves. Your model doesn’t. Hallucinations multiply. Context fades. Reliability crumbles. Solution RAG. Retrieval-Augmented Generation isn’t a trick. It’s the system that makes LLMs think before they speak. It pulls live knowledge. It adapts instantly. It scales without retraining. From the Mastering RAG playbook: RAG is architecture, not a feature. It separates retrieval from reasoning. It turns static LLMs into living, learning systems. It’s how Anthropic, Cohere, and Databricks ship smarter AI fast. Fine-tuning got us here. RAG will take us further. ➕ Follow Ghadeer for more insights ♻️ Repost to help others in your network 📩 Save for later

252

79 commenti

Hi All Snowflake announced Cortex Code this week at BUILD London, and I’ve been testing it with real enterprise environments. My impression is that this isn’t another “AI writes SQL faster” tool! What surprised me is that it actually operates inside your Snowflake context: 🦾 understands schemas, roles, masking policies 🦾 reasons about query history and warehouse cost 🦾 respects governance instead of hallucinating optimizations Example questions that worked out of the box: “Why is this warehouse expensive?” “Which tables with PII are queried most often?” “Refactor this dbt model but keep our naming conventions” 💥 This can really help your teams ship production pipelines super fast, especially for messy legacy setups! That said, there’s a real cost/governance tradeoff here that Snowflake doesn’t fully solve yet, especially once you start scaling Cortex usage itself. I wrote a deeper breakdown here, what works, what doesn’t, and where costs sneak up: https://lnkd.in/dhErFNt4 thanks to the following Snowflake Super Heroes (not all of you know me but I love reading your blogs!) for your inspiring blogs and discussions -> John Ryan Sean Falconer Tim Spreitzer Tomáš Sobotík Snowflake Seemore Data #datacloud #data #dataengineering #seemoredata

44

8 commenti

Quick note. Looking at Data Scientist interview activity, it has actually dropped over 50% year on year, but the work hasn't gone anywhere, it's just moved into titles like ML Engineer and AI Engineer instead. So if you're still hiring against an old-school "Data Scientist" spec, you might be fishing in a shrinking pool for a role that's quietly split into three different jobs. Nearly half of all data and analytics postings now mention AI somewhere in the spec. That's not a small trend, that's the market creating a new "normal" looks like for the role. My clients and partners want people who can actually ship something into production, not just build a model or a dashboard and hand it off. Domain depth in a specific industry is mattering more too, generalists are having a tougher time standing out than they were a year ago. If you're hiring or job hunting in this space right now, worth checking whether your spec or your CV is speaking last year's language. Feel free to get in touch for any value adds!

139

9 commenti

Good post from the team behind Apache Sedona—especially the work from Jia Y., James Willis and Pranav Toggi introducing SpatialBench. The discussion around spatial performance lately has focused on speed claims (like the recent Serverless SQL comparison from Databricks), but what stands out here is the focus on price-performance and query coverage, not just runtime. Using SpatialBench, the results showed that while Serverless SQL performed well on one of the simpler queries, Sedona delivered up to ~6× better price-performance across several others and supported more of the benchmarked spatial workloads. My takeaway: benchmarking spatial systems is tricky, especially when serverless layers hide the actual compute. Normalizing for cost and testing across different query complexities gives a much clearer picture. Worth a read if you’re working with large-scale geospatial analytics. https://lnkd.in/gAXb55md

26

Coalesce is working hand in hand with Snowflake to get data engineers more efficient than ever. Data engineering teams are under pressure to deliver more. Faster. More data products for analytics and AI. Faster than ever but without compromising governance or quality. Too much time is still spent stitching together infrastructure, maintaining brittle pipelines, and firefighting issues instead of designing trusted, reusable data models. While most enterprises focus their AI efforts on dashboards, copilots, and applications, a quiet AI revolution is reshaping how data engineering work itself gets done. Join the webinar with Snowflake and Coalesce.io to see how AI can help your data teams move faster, keep tighter control over data quality, and deliver more reliable data products to the business. You'll see: - Panel Interview from Mark van der Heijden and Jenny Z. - Short demo of Snowflake: see how AI in Snowflake can speed up key data engineering tasks - Short demo of Coalesce: watch how Coalesce streamlines the design, automation, and governance of data transformations   Link to register: https://lnkd.in/eYhRkQef

22

I just watched a very interesting bauplan webinar on data quality with Claude Code https://lnkd.in/e6g9TM7U I highly recommend giving it a watch, a couple of things stood out to me: - the ability to use GenAI on actual data within a developer workflow, rather than in isolated chat tools - how the entire workflow stays within an IDE and terminal, including branch management

8

AI Tech Stack Funding this week: $5.6bn Eleven companies that raised this week in the AI infrastructure / tool stack space: CoreWeave – Post-IPO Equity – $2bn: A cloud-based AI infrastructure company offering GPU cloud services to support large-scale AI and machine learning workloads. StepFun – Series B – $718.1m: An artificial intelligence company focused on advancing toward artificial general intelligence (AGI). AstraSync AI – Seed – Undisclosed: Know Your Agent platform providing identity, trust, and verification for the agentic economy. Databricks – Debt Financing – $1.8bn: A data and AI platform unifying data engineering, analytics, and machine learning on a lakehouse architecture. LiveKit – Series C – $100m: A cloud platform enabling developers to build, deploy, and manage real-time communication and AI-driven applications. AppliedAI – Series B – Undisclosed: Provides AI-augmented automation for mission-critical business workflows. Upscale AI – Series A – $200m: Redefines networking with purpose-built solutions for high-performance compute environments. Baseten – Venture – $300m: An AI infrastructure platform integrating machine learning into production business systems and workflows. humans& – Seed – $480m: A frontier AI research and product lab focused on human-centred AI systems and interaction models. Thunder Compute – Seed – $4.5m: One-click GPU instances designed to deliver compute at significantly lower cost. NENNA.AI – Seed – Undisclosed: An AI enablement platform focused on the secure application of AI technologies. #AI #ArtificialIntelligence #AITechStack #AIFunding #AIEcosystem #AIInfrastructure #MachineLearning #ML #Startups #TechFunding #VentureCapital

6