Activity
7K followers
-
Molham Aref reposted thisMolham Aref reposted this𝘛𝘰𝘸𝘢𝘳𝘥𝘴 "𝗕𝗲𝘀𝘁 𝗶𝗻 𝗧𝗲𝗹𝗰𝗼 𝗟𝗟𝗠𝘀" Yesterday we announced, in conjunction with GSMA, the release of the open telco models, 30 models of various sizes and architectures, which were built to address four things: 1. 𝗔𝗰𝗰𝘂𝗿𝗮𝗰𝘆: While frontier LLMs (e.g., OpenAI, Anthropic) have made tremendous advances in solving coding and math in recent years, their understanding of telco is still only around 70% accurate. Our goal was to improve accuracy from 70% to 90%+ (which we did). 2. 𝗖𝗼𝘀𝘁 𝗮𝗻𝗱 𝗹𝗮𝘁𝗲𝗻𝗰𝘆: If one wanted to run frontier models at network scale for optimization or troubleshooting, it would be very costly. Smaller fine-tuned models, on the other hand, are typically much more cost-effective and faster. 3. 𝗢𝗻-𝗱𝗲𝘃𝗶𝗰𝗲 / 𝗼𝗳𝗳𝗹𝗶𝗻𝗲 𝗰𝗮𝗽𝗮𝗯𝗶𝗹𝗶𝘁𝘆: There are times when the network is not available (e.g., an outage), and it would be beneficial if we could utilize an on-device Small Language Model (SLM) to help retrieve information and answer questions. 𝟰. 𝗙𝗹𝗲𝘅𝗶𝗯𝗹𝗲 𝗼𝗽𝗲𝗻-𝘀𝗼𝘂𝗿𝗰𝗲 𝗳𝗿𝗮𝗺𝗲𝘄𝗼𝗿𝗸 (𝗚𝗣𝗨-𝗮𝗴𝗻𝗼𝘀𝘁𝗶𝗰): To keep the GPU industry competitive, we wanted to train many different models (30+) with relatively new open source GPU-agnostic software #ScalarLM (supported by RelationalAI) that works on both NVIDIA and AMD. This work was accomplished thanks for hundreds of the latest AMD GPU's hosted by TensorWave, and the models, and training software is all open-sourced. Overall, this work would have been very difficult and costly if it were not for the millions of pages of data/documents provided by GSMA members, and the great work and contributors described by Farbod Tavakkoli who along with Gregory Diamos, and Ashish Vaswani will hold a Masterclass on this Weds. Thanks also to Louis Powell for facilitating this work as part of the Open Telco effort, and our AT&T colleagues supporting the work, Roderic Paulk, Gus Mustakas, Jorden Terrazas, Abhay Dabholkar. read more here on Farbod's post: https://lnkd.in/g_B3QVkK
-
Molham Aref shared thisThank you Gartner for publishing the first ever MQ on Decision Intelligence and making the case for this important category. We are proud to be included. If running your business is like driving a car, Business Intelligence is like the dashboard: it tells you how fast you are going, how much fuel you have left, and how many miles you have travelled. Decision Intelligence is the navigation system, telling you how to best get to your desired destination. In our view DI builds on BI. They are not different stacks. Businesses need both.Molham Aref shared thisWe’re excited to announce that RelationalAI was included in the first ever Gartner® Magic Quadrant™ for Decision Intelligence Platforms, a mere 13 months after the general availability of our solution as a Snowflake Native Application! RelationalAI—evaluated alongside 16 other vendors— has the only platform built as a Snowflake Data Cloud-native Decision Intelligence Platform in this assessment. As Carlie Idoine wrote in the Gartner announcement, “Critical business decision models cannot be affordably managed across disconnected platforms and processes.” We are proud to have deeply integrated our Decision Intelligence capabilities with the leading platform for Business Intelligence, bringing together DI and BI platforms for the first time. To access the full complimentary report, click on the link in comments 👇
-
Molham Aref shared thisI’m delighted to announce that Snowflake and AT&T have invested in RelationalAI. Nothing beats the validation of your customers and partners becoming shareholders. Thank you Andy Markus, Mark Austin and the team at AT&T. Our joint work on specializing LLMs for each of your data and semantic estates is accelerating the way enterprises create value with AI in the Enterprise. Thank you also Sridhar Ramaswamy, Christian Kleinerman, and the Snowflake team. You have built the best and most widely adopted AI Data Cloud and made it possible for RelationalAI and others to deeply integrate. Together we are building the best platform for AI powered Decision Intelligence.Molham Aref shared thisSnowflake Ventures and AT&T Ventures have invested $22.5M in RelationalAI to accelerate the next stage of enterprise decision intelligence natively inside the Snowflake AI Data Cloud. This marks an important milestone for RelationalAI and for the evolution of enterprise AI. Enterprises have been searching for AI that can support real decisions — not just chat, summarize, or predict, but deliver outcomes grounded in their own data, logic, rules, and governance. Today’s frontier models don’t have that visibility, which keeps GenAI at the edges of day-to-day decision-making. RelationalAI closes this gap. Running entirely inside Snowflake, our system brings reasoning, semantics, and enterprise-specific LLM training to where data already lives. The result is fast, explainable, decision-grade intelligence delivered without data movement and without compromising trust. With support from Snowflake Ventures and AT&T Ventures, we’re advancing a future where AI doesn’t just analyze the past, but helps organizations run their business with confidence. Read more below: 👇 https://lnkd.in/gNtjFpXe #Snowflake #ATTVentures #DecisionIntelligence #EnterpriseAI #AIDataCloud #RelationalAIRelationalAI launches GenAI-powered Decision Intelligence system on the Snowflake AI Data Cloud – RelationalAIRelationalAI launches GenAI-powered Decision Intelligence system on the Snowflake AI Data Cloud – RelationalAI
-
Molham Aref shared thisWe're delighted to partner with you Jay Hinton and the team at BNYMolham Aref shared thisWhat does Decision Intelligence look like inside a 240-year-old financial institution trusted with trillions of dollars? BNY's Global Head of Product for Data and Analytics, Jay Hinton, joined our own Siddhartha Bala to talk about how they're reshaping decisions inside the business with RelationalAI, running natively inside Snowflake. Tune in to learn about: • Moving from data-centric to decision-centric operations. BNY is building systems that don’t just report the past, but help the business understand what comes next. • AI that stays inside the enterprise. Governed, explainable, and aligned to existing controls. • A shared semantic model that reflects how the business actually works. This is what makes predictions, rules, and optimization consistent across teams. • A safer way to bring advanced reasoning to high-stakes environments. With RelationalAI, reasoning runs where the data already lives: Snowflake. Watch the full conversation in the link in comments ⬇ #FSI #BNY #SnowflakeNative #RelationalAI #AIDataCloud“We’re evolving from data-centric to decision-centric operations.” — Jay Hinton, BNY“We’re evolving from data-centric to decision-centric operations.” — Jay Hinton, BNY
-
Molham Aref shared thisOnly 20% of the world’s data is public and there is consensus that it’s tapped out, resulting in a plateauing of LLM scaling. Frontier model providers like Anthropic and OpenAI are trying to work around this limitation in a variety of ways, including the generation of synthetic data. We’re taking a different approach that makes sense in an enterprise setting. We use the enterprises own private data to keep the scaling laws going. We have a breakthrough for efficiently converting private relational data into tokens -and- we have an asymptotically faster approach for RL based fine-tuning. We use the Spider 2.0 benchmark to demonstrate predictable scaling on relational workloads. We are able to cost effectively fine tune an open LLM for each of the ~150 Spider 2.0 databases — improving accuracy and reasoning power deterministically as we increase compute. The scaling laws live-on in the Enterprise! You can read more about our approach on our blog announcing Superalignment for Relational Databases: https://lnkd.in/g4hhenpk Thanks to Superalignment, I’m happy to share that we’re back to the top of the leaderboard on the Spider 2.0 benchmark. Thank you to our partners at AT&T for their brilliance, to TensorWave and AMD for providing us with screaming fast GPUs, to the Spider 2.0 team for their awesome benchmark that helps advance the state of art in text-to-SQL research, and to Snowflake for hosting the Spider 2.0 benchmark datasets. Andy Markus, Mark Austin, Abhay Dabholkar, Farbod Tavakkoli, Gregory Diamos, Naila Farooqui, Sudnya Diamos, Rich Schiavi, Dali Ben Smida, Nikolaos Vasiloglou, Pigi Kouki, RelationalAIIntroducing Superalignment for Relational Databases – RelationalAIIntroducing Superalignment for Relational Databases – RelationalAI
-
Molham Aref reposted thisMolham Aref reposted thisThe future of enterprise AI isn’t just about smarter models, it’s about building a common language between humans and machines. In his recent interview with InfoQ, RelationalAI's VP of Research ML, Nikolaos Vasiloglou, describes this shift: “Knowledge graph is a language that both humans and machines understand...and that’s the glue.” Nik goes on to explain what many companies are still missing - modern applications are no longer backed by traditional databases, they're backed by knowledge graphs. “We’re seeing the rise of language-model–backed modern applications with the rise of knowledge graphs." "They go together. They enable GenAI." "They enable enterprise applications.” Listen to the full podcast and read the transcript via the link below. s / o Srini Penchikala
-
Molham Aref reposted thisMolham Aref reposted thisWrapping up another amazing Snowflake Summit. It was great to connect with so many customers, colleagues and friends. And to share some of the latest product updates from RelationalAI. This week we announced ... > Text2Reasoner capabilities > semantic view interoperability > prescriptive reasoning > new graph reasoning capabilities > predictive (GNN based) reasoning You can read more about our product updates here - https://lnkd.in/g6JS-RHH - and get started with RelationalAI on the Snowflake Marketplace. We also had an opportunity to share some of the cool work we are doing with AT&T that has delivered top of the leaderboard results for the Spider 2.0 text2SQL benchmark as of May 30, 2025 - https://lnkd.in/gmtcAPJ6Unlock Enterprise AI with Relational Intelligence for the Snowflake Cloud · RelationalAIUnlock Enterprise AI with Relational Intelligence for the Snowflake Cloud · RelationalAI
-
Molham Aref reposted thisMolham Aref reposted thisThe foundational pillar of our AI work at AT&T is being great at AI to drive business value. In the era of Generative AI and now Agentic AI, one of the most critical aspects of being truly world class with AI is to optimize the business value and subsequent return on the investment with the ability to turn human language into computer language to execute complex analytics - as so many of our GenAI and Agentic AI use cases involve some type of data functionality. AT&T is still #1 in the world on the core BIRD Generative AI SQL Benchmark (more news coming there soon), but I’m so thrilled to announce that we’ve now achieved the top spot in the world on the Spider 2.0 benchmark! Spider 2.0 is a rigorous benchmark for evaluating how well AI systems can understand natural language and generate correct SQL queries over complex, real world databases. We’ve partnered with RelationalAI (RAI) on the submission combining Generative AI along with relational knowledge graph technology to achieve the results. As I repeatedly say, just because you can do Generative AI doesn’t mean you do it well. These benchmarks are so important as they show AT&T’s approach is empirically world class and provide the most accurate data results to drive the expected business value. Thank you to the Spider 2.0 team for maintaining a world-class benchmark that pushes the entire field forward and a shout-out to our incredible partners at RAI!
-
Molham Aref shared thisProud moment for us at RelationalAI. Thank you AT&T CDO for your partnership. Andy Markus Mark Austin Abhay Dabholkar Farbod Tavakkoli Matt Dugan.Molham Aref shared thisWe’re thrilled to share that in collaboration with our partners at the AT&T CDO office, we just achieved a new State-of-the-Art (SOTA) result on the Spider 2.0 benchmark for question answering with Generative AI via text-to-SQL. Spider 2.0 is one of the most rigorous benchmarks for evaluating how well AI systems can understand natural language and generate correct SQL queries over complex, real world databases. Our joint submission is at the top of the leaderboard as of today. This milestone is now reflected on the official leaderboard at https://lnkd.in/eMVdj-vV Big thanks to the Spider 2.0 team for maintaining a world-class benchmark that pushes the entire field forward. Interested in learning more about how AT&T combines generative models with knowledge graphs? Our very own Nikolaos Vasiloglou will be presenting with Farbod Tavakkoli from AT&T on better AI accuracy in an exclusive breakout session during Snowflake Summit: 🗓️ Tuesday, June 3 🕒 03:00 PM - 03:45 PM 📍Room 209 (Moscone South, Level 2) ❄️ We’re just getting started. Onward! 🚀
Molham Aref commented on a post
1mo
Delighted to join you Mark Austin to talk about the work our teams are doing together.
Molham Aref commented on a post
1mo
It was great to be back on theCube with David Vellante and John Furrier at MWC this week. Thanks Mark Austin for a fun conversation about the industry leading AI innovations led by AT&T.
Molham Aref commented on a post
1mo
Congratulations Farbod Tavakkoli, Mark Austin, Andy Markus, the teams at AT&T, AMD, GSMA and everyone else involved. We are happy for RelationalAI to have contributed the open training and inference stack used to build the OTel family of models.
-
Molham Aref liked thisMolham Aref liked thisReasoning in RelationalAI is grounded in your ontology, so you can define and solve different problems – and integrate their solutions – on a shared base. To get you started, we have 20+ example templates covering optimization use cases like: - Energy grid interconnection planning - Machine maintenance scheduling - Supply chain reliability Each template is a jumping-off point: adapt the data, adjust the formulation, and solve within your own Snowflake warehouse. Browse at https://lnkd.in/gJxTCbFM
-
Molham Aref liked thisMolham Aref liked thisEveryone's racing to build AI data centers, but the grid can't keep up. Watch an agent powered by RelationalAI forecast demand → check feasibility → analyze criticality → optimize data center approvals and substation upgrades per investment scenario. It's all grounded in the ontology and powered by structured reasoning, each step building on the last. https://lnkd.in/gsTBEg4k
-
Molham Aref liked thisMolham Aref liked thisScalarLM 1.0 blog - https://lnkd.in/gzzcp8TK
-
Molham Aref liked thisMolham Aref liked thisMost companies have invested in an AI data cloud, but still struggle to turn data into actual decisions. BI tools answer “What happened?”, but they rarely answer “What will happen?” or “What should we do?” That’s the gap we’re working to close. At RelationalAI, we unlock decision intelligence in Snowflake by augmenting the semantic layer with business ontologies and advanced reasoners. This enables teams to: • Ask questions in natural language about core business concepts, relationships, and rules • Apply advanced reasoning to drive better decisions and business outcomes • Do it all without moving data out of Snowflake On March 17 (10 AM GMT), I'll join Fawad A. Qureshi from Snowflake to showcase a demo where we’ll show this in action. If you’re interested in learning more, register here: https://lnkd.in/gCTcnR6mFrom Data Chaos to Unifying Data, Business Logic and Decisions | 17 MarchFrom Data Chaos to Unifying Data, Business Logic and Decisions | 17 March
-
Molham Aref liked thisMolham Aref liked thisDelighted and proud to share my next adventure: I am stepping into the role of Chief Executive ANZ and taking on the Global Communications Industry Lead responsibility at NCS Group. This is an exciting new chapter — professionally and personally. I will be moving to Australia, very energised about what lies ahead. For NCS, ANZ is a strategic growth frontier. With our strong Australian roots, deep technology expertise and collaborative, can-do culture, combined with our pan-APAC scale and AI-powered solutions, I am hugely energised to work with the talented ANZ team and build something truly special for our clients — spanning Communications, Financial Services, Government, Healthcare and beyond. A thrilling first for me! As Global Communications Industry Leader, I continue driving our business in a sector I have been deeply passionate about for 35 years. The work we have done together with our Telco clients and the results we have achieved, have been inspiring — I am much looking forward to go even further together. Deeply grateful to our new CEO Sam Liew for this trust, and inspired by his vision to rewire NCS with deeper industry expertise, scaled AI capabilities and a client-first culture built for speed, collaboration and maximum impact. To the ANZ team, clients and partners: I am coming with full commitment and great excitement to build something extraordinary together! To my NCS Communications colleagues and Telco clients: very proud and thankful for our cooperation and what we have built and achieved together over the past 12 months — and fully committed and energised to achieving even bigger and better results. Do reach out at tom.loozen@ncs.com.sg to explore how we can jointly create strong business impact — in ANZ or across APAC. #NCS #NCSAustralia #ANZ #Australia #Communications #AI #TechServices #NextHorizon
-
Molham Aref liked thisMolham Aref liked thisWhy is poor liquidity a serious liability? The answer may seem obvious, but as with most things, you have to dig a little deeper to reveal the truth. On a recent episode of our Liquidity Talks live series, we invited Herve Carrere, CPTO Treasury and Capital Market at Finastra to join our host, Neville Roberts, Planixs CEO. During the live episode we talked about AI and real-time intraday liquidity optimisation, focusing on: ✅ Leveraging AI and data aggregation for predictive liquidity management ✅ Payment sequencing, prioritization, and anomaly detection ✅ New operating models for continuous settlement ✅ Tokenized assets and programmable liquidity: opportunities and challenges You can watch the full episode on-demand by following the link in the comments below.
-
Molham Aref liked thisMolham Aref liked thisOn my way back from ECIR 2026 Industry Day (https://lnkd.in/dq2ksqZa) where I had the pleasure to give a keynote on advances in Foundation Models on Business Data. The current generation of such foundation models is capable of in context learning - which requires context to be retrieved accurately and efficiently at inference time, a great point of connection between the ML and the IR community! If you are curious about the details, check out: * ConTextTab: A Semantics-Aware Tabular In-Context Learner - https://lnkd.in/dtZPzY9e * Position Paper on Foundation Models on Business Data - https://lnkd.in/e9mzQ25m Thanks Vinay Setty for the invitation and the organization of the Industry Day!Position: Foundation Models for Tabular Data within Systemic Contexts Need GroundingPosition: Foundation Models for Tabular Data within Systemic Contexts Need Grounding
-
Molham Aref liked thisMolham Aref liked thisMultiple reasoning approaches surface risks no single analysis catches. Supply chain data in Snowflake → RelationalAI ontology → chained reasoning: graph analysis found bottleneck sites, rules flagged high-risk suppliers, optimization routed inventory to minimize cost to meet demand. Then a what-if disabled a "safe" supplier – only to reveal a hidden point of failure. These questions require more than SQL and LLMs. The ontology made this fast – and kept the agent grounded in real business logic and structured reasoning instead of... creative assumptions. https://lnkd.in/gdbKbG7PTransforming Supply Chains with Advanced Insights from RelationalAITransforming Supply Chains with Advanced Insights from RelationalAI
-
Molham Aref liked thisMolham Aref liked thisYou know your AI company needs a real GTM operator when your tenth customer closes completely differently than your first three. The first few? Founder-led. Hand-held. You convinced the prospects because they believed in you, not because the motion was repeatable. By customer ten, if you're still explaining the same things, answering the same objections, customizing the same workflows, you don't have a go-to-market problem. You have a product-market fit problem. But if a pattern is emerging and you're still the only one who can close it? That's when you need someone who's done this before. Someone who can take what's working, codify it, and scale it without you in every conversation. The gap between 10 customers and 100 isn't about working harder. It's about building the machine that works without you. Most founders wait too long to make this hire. But the best GTM operators don't need a playbook handed to them. They build it with you. #ai #leadership #gtm #founders
Other similar profiles
Explore more posts
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content