Emil Eifrem
Menlo Park, California, United States
9K followers
500+ connections
Articles by Emil
-
Ontologies, Semantics, Agents and Knowledge
Ontologies, Semantics, Agents and Knowledge
A few weeks ago I was in San Francisco for the AI Engineer World's Fair organized by swyx . I gave a talk on a topic…
10 Comments
Activity
9K followers
-
Emil Eifrem reposted thisEmil Eifrem reposted thisI'm excited to co-chair a first AI Memory Working Dinner with Kwiri Yang from the Quantum AI Institute, Brian J. Baumann from NYSE, and John Furrier from SiliconANGLE & theCUBE, as Technical Chair. Participants include more than 20 AI leaders from across multiple layers of the AI stack, spanning compute to foundation models, graph and context, memory, orchestration, application, and trust. The two topics we are tackling first are: 1. Agentic Identity, Authorization, and Memory Access: Who is an agent acting on behalf of? What is it allowed to remember, retrieve, change, or share? How do we ensure that consent and permissions remain enforceable as memory passes between agents? 2. Persistence, Portability, and Interoperability: How does memory survive beyond a single interaction and move safely across agents, models, systems, and vendors? What shared protocols are needed to preserve its meaning, provenance, and access controls? My co-facilitators include: Ali Tore, who leads Conversational AI at RingCentral; Arun Rao, AI leader at NVIDIA on Nemotron, and machine learning practitioner and Adjunct Professor at UCLA; Vasilije Markovic, Founder and CEO of cognee; Lake Dai, Founder and Managing Partner of Sancus Ventures and Adjunct Professor at Carnegie Mellon University, bringing expertise in AI deployment, infrastructure, and governance. Our goal is to understand what companies are encountering in practice, identify where existing approaches fall short, and agree on the problems we can meaningfully tackle together. Looking forward to sharing our journey and learnings as we progress!
-
Emil Eifrem shared thisWait Kerstin Eifrém, how did I end up in a universe where the New York Stock Exchange is posting about my love for, well, Britney Spears? 🤔🤷♂️Emil Eifrem shared thisAI is only as smart as the connections in your data 🚀⚡️ From Britney Spears to AI 🤯, Emil Eifrem, CEO & co-founder of Neo4j, joins theCUBE + NYSE Wired studio for a conversation with his longtime friend and tech industry veteran John Furrier. Emil breaks down why the real power of AI isn’t just the data you have, but how that data connects. Full interview 👇 https://lnkd.in/g_tDKn7N John Furrier | Gemma Allen | Brian J. Baumann #NYSEWired #theCUBE #Neo4j
-
Emil Eifrem reposted thisEmil Eifrem reposted thisGreat energy in NYC this week at Graph Summit! 🚀 Graphs are becoming the foundational knowledge layer that makes AI trustworthy and actionable. Key takeaways from the customer and keynote sessions: 1. Neeraj Deshmukh(Blitzy): AI is a map, and a map is a graph. Unlocking infinite context—whether in code or enterprise data—requires structural mapping. 2. Harsh Acharya (Dell Technologies) : Stop trying to simply automate existing processes. Instead, redefine your processes entirely to native design for the agentic, autonomous era. 3. Emil Eifrem (Neo4j): For AI agents to act reliably at scale across disparate data, they need three core pillars: Meaning (Ontologies), Reach (Multi-hop traversal), Learning (Memory) Incredible insights from industry leaders, enterprise practitioners, and the Neo4j team. Thanks Drew Palsgrove, Kerry Khoo-Fazari, PhD Engineering, Felix Van de Maele for joining us. #GraphSummit #Neo4j #AI #AgenticAI #KnowledgeGraph #TechLeadership #DataArchitecture
-
Emil Eifrem shared thisWhat a great set of links! Missed it on Friday but I can verify that it worked well with my Monday morning coffee too. ☕Emil Eifrem shared thisThree things worth ten minutes with your Friday coffee. 1. A Neo4j semantic layer over BigQuery and Databricks cut Text2SQL token use by up to 81% on a catalog of 278 near-identical Census tables. On Databricks, it took one model from $1,527 to $693 a month at 10K queries, and every model tested landed 11/11. Accuracy moved too: a cheap model went from 2/4 to 4/4 once the schema carried business meaning instead of column names. https://lnkd.in/d-ss8X8E 2. Going Meta wrapped a 10-episode season on building and grading ontologies. Episode 4 turns "is this ontology any good" into measurable checks across four dimensions, including how well it answers its competency questions, shipped as Cypher queries and a dashboard on GitHub. Competency questions as acceptance tests is the exact thing I wrote about last month, so seeing it land as a dashboard was a good Friday. https://lnkd.in/dyKt5SHz -> https://lnkd.in/dznCtX6T 3. A paper from 10 September separates where Graph-RAG answers actually break. Structural defects in the graph dropped accuracy from 0.93 to 0.39. The model then fills broken paths with what it learned in training, which hides more than 70% of the query errors you would otherwise catch. Your monitoring stays green while the data rots underneath. https://lnkd.in/dGhZ4JTu #Neo4j #KnowledgeGraphs #GraphRAG #GraphDatabase
-
Emil Eifrem reposted thisEmil Eifrem reposted thisReally excited about this one. Today, we launched "Neo4j GraphAware Financial Crime Intelligence". I have spent a lot of time working in financial crime, from real-time fraud detection at Barclays, to working with key financial crime vendors at Elastic, and now five years at Neo4j, helping banks see the networks behind the numbers. This is the Neo4j product I have wanted to present to investigators for a long time. I am really excited about what the combination of Neo4j and GraphAware can do. This is only the start. The idea is simple. Stop giving investigators isolated flags and start giving them connected context. One open graph native foundation across the whole financial crime estate, built on a knowledge layer that pulls internal data, third-party intelligence, and detection signals into a single, entity-resolved view. 🔹 Signal: detect patterns across connections, not just rows 🔹 Alert: every alert arrives with its full context 🔹 Investigate: follow the evidence wherever it leads, no query required 🔹 Decide: capture the evidence and produce a defensible, explainable report "Every fraud involves a network. Every network has a pattern. This is how you find it." Huge thanks to the GraphAware and Neo4j teams who got this over the line. #FinancialCrime #FraudDetection #Neo4j
-
Emil Eifrem shared thisIn just a few days in NYC!Emil Eifrem shared thisAnnouncing #GraphSummit NYC! 🍎 Thrilled to share about our new GraphSummit edition in NYC this September 17, 2026. Request to attend today, as space is limited: you'll hear from #Neo4j experts and AI leaders as they share how they're building enterprise #AI with knowledge layers and bringing graph-powered AI into production. In the afternoon, you'll have the chance to attend a hands-on GenAI workshop, where you'll build AI applications with our technical experts. Join Sudhir Hasbe, Jesús Barrasa, and more Neo4j experts, along with customers from Pfizer, Dell Technologies, Fidelity, Kerry Khoo-Fazari, PhD Engineering from RBC, and Felix Van de Maele from Colibra. See you! https://bit.ly/3SpKDVO #enterpriseknowledgelayer #AI #graphdatabase
-
Emil Eifrem shared thisGreat summary of a great article!Emil Eifrem shared thisKarpathy's LLM wiki recipe is smart. Markdown on a file system is not. Grep reads every file linearly. Cross-links cost another full scan per hop. Backlinks have no index. It degrades quietly as the corpus grows. Zach Blumenfeld's fix: replace the file system with a Neo4j graph. A hop is a pointer, not a scan. Zero tokens at index time. NICD research shows 2x precision and recall and 80% improvement in truthfulness versus vector-only retrieval. He built ki to show it. https://bit.ly/4ylrBza
-
Emil Eifrem shared thisThe best part of building Neo4j hasn't changed since day one: seeing what customers build with graphs that we never would have imagined ourselves. Two years ago most of them were experimenting with graphs and AI. Now they're running multiple graph-backed AI applications and agents in production, and they're coming to GraphSummit NYC to learn from each other. That's what I'm most looking forward to. Harsh Acharya from Dell Technologies, Krishna V. from Fidelity Investments and Kerry Khoo-Fazari, PhD Engineering from RBC and many more will share what they're building and what they've learned along the way. And they will be joined by the Neo4j experts from the frontier of graphs & AI like Jesús Barrasa and Sudhir Hasbe. I'll open the day with where I think all of this is heading: the knowledge layer for trustworthy AI. Then I'm going to sit down and listen. If you're in New York on September 17, come join us. Request your seat: https://lnkd.in/dJRq9SCr
-
Emil Eifrem reposted thisWe still have a few seats left! And it’s a great opportunity to learn more about what the GraphAware acquisition means for the future of Neo4j in FinServ.Emil Eifrem reposted thisSomething exciting is happening for the financial services community in Stockholm. On September 16, we're bringing together industry leaders and graph experts at At Six for what will be one of this year's most interesting GraphTalks. Following Neo4j's acquisition of GraphAware, we can now offer a more complete Graph Intelligence platform, combining powerful graph technology with intelligence capabilities for mission-critical environments where data sovereignty is essential. September the 16th. Join us and be part of the conversation. https://lnkd.in/d_dgM-Du
-
Emil Eifrem liked thisEmil Eifrem liked thisI'm excited to co-chair a first AI Memory Working Dinner with Kwiri Yang from the Quantum AI Institute, Brian J. Baumann from NYSE, and John Furrier from SiliconANGLE & theCUBE, as Technical Chair. Participants include more than 20 AI leaders from across multiple layers of the AI stack, spanning compute to foundation models, graph and context, memory, orchestration, application, and trust. The two topics we are tackling first are: 1. Agentic Identity, Authorization, and Memory Access: Who is an agent acting on behalf of? What is it allowed to remember, retrieve, change, or share? How do we ensure that consent and permissions remain enforceable as memory passes between agents? 2. Persistence, Portability, and Interoperability: How does memory survive beyond a single interaction and move safely across agents, models, systems, and vendors? What shared protocols are needed to preserve its meaning, provenance, and access controls? My co-facilitators include: Ali Tore, who leads Conversational AI at RingCentral; Arun Rao, AI leader at NVIDIA on Nemotron, and machine learning practitioner and Adjunct Professor at UCLA; Vasilije Markovic, Founder and CEO of cognee; Lake Dai, Founder and Managing Partner of Sancus Ventures and Adjunct Professor at Carnegie Mellon University, bringing expertise in AI deployment, infrastructure, and governance. Our goal is to understand what companies are encountering in practice, identify where existing approaches fall short, and agree on the problems we can meaningfully tackle together. Looking forward to sharing our journey and learnings as we progress!
-
Emil Eifrem liked thisEmil Eifrem liked thisI am honoured to be named to TIME's Executives of the Year: Tech and Data list. Thank you, TIME, for the recognition of Citi's AI journey. This is the work of thousands of colleagues across the company who are helping turn the potential of AI into real-world capabilities for our clients and our firm, and I am proud to be part of this team. AI is creating new possibilities around the world. Citi is helping lead how these capabilities are applied responsibly and at scale across financial services. https://lnkd.in/edeYjdHrDavid Griffiths Is one of TIME's 2026 Executives of the Year: Tech and DataDavid Griffiths Is one of TIME's 2026 Executives of the Year: Tech and Data
-
Emil Eifrem liked thisAfter years of researching how CEOs communicate strategy, I can honestly say that this conversation with Neo4j founder and CEO Emil Eifrem may be the best "popular pitch" for #strategy I have ever seen. Clear, compelling, and deeply resonant it gets the message across. Just clicks with me, and makes a lot of SENSE. Judge for yourself.Emil Eifrem liked thisAI is only as smart as the connections in your data 🚀⚡️ From Britney Spears to AI 🤯, Emil Eifrem, CEO & co-founder of Neo4j, joins theCUBE + NYSE Wired studio for a conversation with his longtime friend and tech industry veteran John Furrier. Emil breaks down why the real power of AI isn’t just the data you have, but how that data connects. Full interview 👇 https://lnkd.in/g_tDKn7N John Furrier | Gemma Allen | Brian J. Baumann #NYSEWired #theCUBE #Neo4j
-
Emil Eifrem liked thisEmil Eifrem liked thisAI is only as smart as the connections in your data 🚀⚡️ From Britney Spears to AI 🤯, Emil Eifrem, CEO & co-founder of Neo4j, joins theCUBE + NYSE Wired studio for a conversation with his longtime friend and tech industry veteran John Furrier. Emil breaks down why the real power of AI isn’t just the data you have, but how that data connects. Full interview 👇 https://lnkd.in/g_tDKn7N John Furrier | Gemma Allen | Brian J. Baumann #NYSEWired #theCUBE #Neo4j
-
Emil Eifrem liked thisEmil Eifrem liked thisGraphSummit NYC is in the books, and what a room it was 🚀 Emil Eifrem opened with the knowledge layer for trustworthy AI. Sudhir Hasbe walked through where the Neo4j Graph Intelligence Platform is headed next. And in between: a new financial crime intelligence offering with GraphAware, a fireside chat with Dell Technologies, and a customer panel where Pfizer, RBC, and Blitzy talked candidly about what it actually takes to build a knowledge layer at scale. The day didn't stop at the main stage. In the afternoon, folks got hands-on in our GraphRAG workshop and dug into strategy in the Executive Program, featuring sessions by Felix Van de Maele of Collibra, a fireside chat between Maribel Lopez and Kerry Khoo-Fazari, PhD Engineering of RBC, and an interactive discussion with Jesús Barrasa and Bryan Nairn. None of this happens without the people who showed up to share it. Thank you to our partners, speakers, and the community who made GraphSummit NYC 2026 unforgettable! #GraphSummit2026 #neo4j
Experience
Other similar profiles
Explore more posts
-
Dan Goosewin
Very smart design decision by the Mastra team. Composite storage acknowledges a reality in AI agents engineering that many systems ignore: different data domains have very different performance, durability, and cost profiles. Forcing everything into one datastore is convenient early on, but it does not scale. Letting memory, workflows, and telemetry live where they work best, behind a single configuration, is a clean abstraction and a meaningful step toward production-grade agent systems.
4 Comments -
Anna Widenius
Open source is easy to talk about. Much harder to sustain. What we are seeing now in #MariaDB is something I find genuinely encouraging: community contributions to MariaDB Server are rising sharply — in some recent months up to 3× compared to last year. Even more important: we are not slowing down under that load. Review and merge times are improving. That combination matters. Because real open source is not just about code being available. It is about people choosing to participate — and the system being able to keep up when they do. This is what a healthy ecosystem looks like: more contributors, faster feedback loops, and shared responsibility. And yes — this is exactly the kind of signal that tells you a project is alive. Full blog: https://lnkd.in/dsdxv-Wd #MariaDB #OpenSource #Database #MySQL #PostgreSQL #DataEngineering #TechLeadership #DigitalSovereignty
2 Comments -
Christophe Pettus
For roughly twenty years, the answer to "can I give the PostgreSQL planner a hint?" has been a firm no. PostgreSQL 19 Beta 1 ships pg_plan_advice, and it's the most serious move toward "yes" the project has made. It is not Oracle- or MySQL-style hints embedded in SQL comments. The mechanism sits outside the query text: you load the module, use EXPLAIN (PLAN_ADVICE) to capture an advice string for a plan you like, and apply that advice externally, keyed by query. The stated goal is to stabilize plans you trust and to experiment with plans the planner currently rejects, without scribbling hints into application code. That distinction matters. Inline hints rot in place and hide from review; externalized advice can be inspected, versioned, and removed. It is also a new tool with sharp edges. A pinned plan is a bet that the data distribution which justified it won't change, and that bet can quietly go bad as the table grows. If you have a query that periodically flips to a terrible plan, or you're weighing whether to pin a plan at all, that's a conversation worth having before you ship it. https://lnkd.in/eHMW4_HW pgexperts.com
1 Comment -
🚀 Zaid Daba'een
OpenAPI 3.2 introduces something subtle but important: event stream support. Until now, OpenAPI specs were built around one-off responses. That worked fine for REST, but not for modern LLM APIs, where responses are streamed token by token. With 3.2, you can now formally describe continuous or event-based responses, not just single payloads. This brings real benefits for developers and toolchains: - Docs and SDKs can render streaming endpoints natively. - Validators can test long-lived responses properly. - LLM providers can finally publish accurate, compliant specs. It’s a small addition with big consequences: OpenAPI is catching up with how APIs actually communicate today - in streams, not snapshots.
1 Comment
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content