Patrick McFadin’s Post

Want a clear signal that lots of people are building AI applications leveraging PostgreSQL? Go look at the GitHub star history for pgvector. Pgvector is a popular PostgreSQL extension designed to store and query vector embeddings, a key component in AI applications like semantic search, RAG pipelines and recommendation systems. Their is a reason why great companies like Timescale, Neon and Supabase all added support for this extension super quickly. Having your embeddings next to your data is a huge advantage from a cost, devx and latency perspective.

Maintain Chat History in Generative AI Apps With Valkey https://ift.tt/2bcHeaO A while back I wrote up a blog post on how to use Redis as a chat history component with LangChain. Since LangChain already had Redis chat history available as a component, it was quite convenient to write a client application. But, that's not the same with langchaingo which is a Go port of LangChain. I am going to walk through how to do the same, but for Valkey (not Redis). Valkey is an open-source alternative to Redis. It's a community-driven, Linux Foundation project created to keep the project available for use and distribution under the open-source Berkeley Software Distribution (BSD) three-clause license after the Redis license changes. via DZone AI/ML Zone https://bit.ly/41qvfZp July 04, 2024 at 04:00PM

Is Apache Flink the secret weapon for batch data processing? 🧐🐿️ This comes up a lot. There's a common misconception that #apacheflink is tailored exclusively for real-time data processing. This is only partially true. Flink is changing data handling with an approach that transcends existing boundaries, offering unmatched efficiency and cost reductions. Making it vital to companies like Netflix, Uber, LinkedIn, and many more. Here’s why we believe Apache Flink is more than a tool; it's a paradigm shift. Old vs. New 🐿 Traditionally, data processing involved operations such as joins only if corresponding datasets are available - either your data got enriched, or it didn’t, based on the query conditions at that moment. 🐿 Flink, on the other hand, innovates by implementing continuous joins. Whether it’s stream or batch data, the process occurs the instant a match is detected. This continuous method enhances data freshness, improves dependency management, and most crucially, fosters both performance and cost optimization. 🐿 In other words, traditional batch joins resemble periodic checks of snapshot states, whereas Flink’s streaming joins continuously build and query state in real time. 🐿 This continuous, flowing process might seem less intuitive but offers profound advantages. Final thoughts #apacheflink is pivoting the processing landscape, pushing the boundaries, beyond real-time data handling. Whether you're already using Flink or considering its adoption, understanding what distinguishes its capabilities can help you manage data operations, making processes more efficient and cost-effective. #apacheflink #streamingdata #realtimedata #datorios

Latency Lowdown: When Data Speed Demons Gossip Tech speed forecast: Brace yourselves for a whirlwind of wisdom as gRPC and Apache Cassandra, the speed demons of data transfer and storage, collide in a high-octane chat! 🚀 In this digital drag race, our tech speedsters rev up about: • The nitty-gritty of HTTP/2 (it's not just alphabet soup, we promise!) 🍜 • Why scaling up is more art than science (hint: it's not just about adding more engines) 🎨 • Memory management magic (or how computers juggle data without dropping the ball) 🤹♂️ • Protocol Buffers: packing data tighter than a tin of sardines 🐟 • Consistent Hashing: the Goldilocks way of data distribution 📊 • Chatting with computer kernels (no, not popcorn kernels!) 🌽 • Keeping your data safe while it's zooming around the digital motorway 🛣️ Caution: High risk of mind-blowing tech jargon and speed-related puns. Bring your geek goggles! 🤓🥽 Tune in now and let the knowledge flow at gigabit speeds! Remember, in our world, the only acceptable lag is in your Netflix stream. 🎬 Ready to accelerate your tech know-how? Zoom into our high-speed broadcast on: 🎵 Spotify 🎙️ Castbox 🔊 SoundCloud 🍎 Apple Podcasts 🎵 Amazon Music ▶️ YouTube Podcasts 🎵 Deezer 🎙️ Podbean 📱 Podcast Addict 🎧 Transistor.fm All these platforms are just a click away at: https://lnkd.in/d8HuEDPZ #Tech #Code #Data #Net #gRPC #DB #Dev #Ops #Sys #API #Cloud #Speed #Scale #Bits #Node #Stack #Hash #Cache #Ping #HTTP #Kern #Mem #Buf #Lag #DistributedSystems #TechTalk #gRPCvsCassandra #LowLatency #TechPodcast #DataTransfer #DigitalOptimization #NetworkPerformance #ITInfrastructure #DataManagement #TechTrends #SystemArchitecture #DevOps #ScalabilityStrategies #EmergingTech #TechLeadership #FutureOfData #PerformanceInsights #TechInnovation #DataSpeedExperts

Thinking of moving your jobs from Spark to Flink? You could do that by changing a single line of code with Apache Wayang. Even more, you could have a mix and match approach to get the best of both. Apache Wayang offers an abstraction for data platform integration together with an optimizer which can automatically select the best platforms for your application. #apachewayang #apachespark #apacheflink #dataplatformintegration

🧠Memory Hackathon with LangChain, https://new.computer/, MongoDB and Anthropic ✨Sign up here: https://lu.ma/taa6ijxt ✨ Long-term memory for LLM applications is one of the most interesting and important - yet also under explored - areas out there. We are THRILLED to be hosting a hackathon on this with an awesome set of partners. 💻 New Computer is building Dot, an intelligent guide, which uses a novel long-term memory system to enable deep personalization. Samantha Whitmore (CEO) was also an early contributor to LangChain 🍥 Anthropic's new Claude-3 models (with long context window and low latency) are perfect for memory intensive applications 🟢 MongoDB's database - with both vectors and general text data - are optimal for exploring different types of memory We will have a great lineup of speakers and hope to make progress in a cutting-edge area. There are limited spots - sign up quick! https://lu.ma/taa6ijxt

We know organizations rely on Kafka for mission-critical systems, and we couldn't bring Bufstream to market without rigorous correctness testing. Together with Jepsen + Antithesis, we've done it, and we're excited to talk about it. We launched Bufstream earlier this year as a drop-in replacement for Kafka. It supports the full #ApacheKafka API, including transactions and exactly-once semantics (EOS). It works with existing clients, stream processing frameworks, Kafka Connect—the whole nine yards. Pre-launch, we partnered with Antithesis to chaos test Bufstream with their deterministic fault injection platform. Combined with Will Wilson's FoundationDB experience, Antithesis was invaluable. Post-launch, we brought in Kyle Kingsbury for his expertise with generative workloads and sophisticated consistency checkers. And that's when things got interesting. During our exploration of Bufstream's behavior, Kyle's tests unearthed a significant flaw in the Apache Kafka protocol itself. Our ability to reproduce this behavior in a deterministic environment is huge. This isn't just about Bufstream anymore—it's about the foundational protocol powering data streaming for countless organizations. The implications could be far-reaching for the entire Kafka ecosystem. Sometimes, you get a chance to uncover issues that impact the entire industry! #DataEngineering #datastreaming #dataops #ai #FutureOfEngineering #TechTrends #Engineering #DevOps #Kafka #distributedsystems #databases

Spark Datatype Casting Issue: Today, we encountered a fascinating issue concerning datatype casting in our real-time pipeline. While consuming data from Kafka, we noticed negative values in an integer column. Upon investigation, we discovered that we were deserializing google.protobuf.Int64Value and then typecasting it into an Int datatype. Interestingly, if the value exceeded the size of the Int datatype, it resulted in a negative value.

Re-creating the bug is usually the most challenging part; listen in as Akshay Shah previews the Jepsen + Antithesis combination we used to demonstrate a transaction bug in the core Kafka protocol reproducibly.

Quickstart Guide to LLM Applications with LangChain & LlamaIndex and MongoDB