--- title: "Real-Time Data Meets Gen AI" newsletter: "MLOps Community" date: 2025-06-26 source: https://aaif.live/newsletters/mlopscommunity/2025-06-26-real-time-data-meets-gen-ai --- # Real-Time Data Meets Gen AI *Plus - benchmarkig intelligence with the ARC Prize, Hidden Gems, ML Confessions, and an upcoming mini summit.* *MLOps Community — Agentic AI Foundation, 2025-06-26* Congrats to our friends Jay and Sammy [https://go.mlops.community/introjun26]! $30M is a bargain if it means never touching brittle ffmpeg and pandas glue code again. Let me know what pain point you'd pay $30M to have fixed. We can't cut that cheque, but we do have access to a ton of smart folks, so maybe we can help anyway. ## Bridging the Gap Between AI and Business Data Violating SLAs, feature store desyncs, lost coffee deliveries - stale data can have disastrous consequences. LLMs won’t deliver much unless they’re wired into the operational data that actually runs the business. Deepti argued that most gen AI apps are still on the sidelines - chatting with PDFs while the useful stuff lives in Postgres, Salesforce, and friends. Her approach flips that by querying source systems in real time, with no pipelines or ETL. Instead of fuzzy guesses, it relies on: * Precise query generation using dialect-specific builders * Live lookups that preserve structure and context * LLMs for routing only, not for constructing SQL This episode actually delivers. Click below to listen. Video [https://go.mlops.community/psl26jun] || Spotify [https://go.mlops.community/ssl26jun] || Apple [https://go.mlops.community/asl26jun] [https://go.mlops.community/asl26jun](https://go.mlops.community/asl26jun) [https://go.mlops.community/psl26jun](https://go.mlops.community/psl26jun) [https://go.mlops.community/ssl26jun](https://go.mlops.community/ssl26jun) ## Hidden Gems ## Practitioner-driven talks and panels, coming to you ## WORLD TOUR The last stop in Cape Town [https://go.mlops.community/wtcapeli] was packed, and had talks on MLOps lessons from a real recommender system, agents in insurance claims workflows, and even what Clippy can teach us about collective intelligence. The tour continues - today hitting Miami [https://go.mlops.community/wtmiami] with teams from Databricks, Google, and Hypermode sharing with a packed room of 150+ engineers how they’re deploying agents in the wild. Next up: * Frankfurt [https://go.mlops.community/wtfrank] - July 3 * Amsterdam [https://go.mlops.community/wtadam] - July 9 * Munich [https://go.mlops.community/wtmunich] - July 16 ## Greg Kamradt: Benchmarking Intelligence | ARC Prize If knowledge is knowing a tomato is a fruit, and wisdom is not putting it in a fruit salad, what’s intelligence? Greg outlined how ARC Prize benchmarks try to answer that, not by checking whether models get the right answer, but by measuring how efficiently they learn - especially compared to humans. Recent results impressed, but raised tough questions about cost, scalability, and tradeoffs: * Low vs high compute: OpenAI’s model scored 75% with low compute but 87% using expensive test-time resources. Performance scales, but not cheaply. * Inference cost matters: Benchmarking now includes latency and energy use, not just accuracy. * Efficiency gap remains: Humans solve these tasks with far less data and energy, highlighting how far models still are from AGI. If intelligence is about learning efficiently, clicking below is a good place to start. Video [https://go.mlops.community/pgk26jun] || Spotify [https://go.mlops.community/sgk26jun] || Apple [https://go.mlops.community/agk26jun] [https://go.mlops.community/agk26jun](https://go.mlops.community/agk26jun) [https://go.mlops.community/pgk26jun](https://go.mlops.community/pgk26jun) [https://go.mlops.community/sgk26jun](https://go.mlops.community/sgk26jun) ## Building AI That Doesn’t Break ## UPCOMING MINI SUMMIT 🗓 Jul 2 | ⏰ 6–7 PM CEST / 9–10 AM PT Flashy demos are easy. Production isn’t. This hour-long session on July 2 is packed with practical ideas for anyone shipping AI in the real world - where APIs fail, humans delay, and YAML grows out of control. * Process Calling: Why stateful agents beat stateless tools * Durable Workflows: Resilient pipelines that recover, retry, and finish the job * No YAML? No Problem: Orchestrate Kubernetes workflows in clean Python For infra folks, AI engineers, and anyone building beyond the notebook. Register now [https://go.mlops.community/msjul2] ## Job of the Week [https://go.mlops.community/jobus26jun](https://go.mlops.community/jobus26jun) [https://go.mlops.community/MLConfess](https://go.mlops.community/MLConfess) Inspired by the podcast: The current approach to RAG and agents is fundamentally broken because it’s built on stale, context-stripped data. Vector stores are the wrong tool for operational data, and ETL pipelines just slow everything down. Harsh or fair? Let me know what you think. Working on something tricky or planning ahead? Here’s how we can help - just hit reply: * Custom workshops tailored to your company’s needs * Hiring? I know some quality folks looking for a new adventure * Want to connect with someone tackling similar problems? I can introduce you Thanks for reading, catch you next time! --- Source: https://aaif.live/newsletters/mlopscommunity/2025-06-26-real-time-data-meets-gen-ai