Protocol Labs’ Post

One of my favorite things about my job is the privilege to interview some of the brightest minds 🧠 Luke Kim and Phillip LeBlanc built Spice AI to provide developers with core building blocks to create data and #AI-driven products, including compute, storage, #zeroknowledge and #machinelearning accelerators, and #blockchains. Read about how these ex-Microsoft engineers built a company that takes the stress out of software expertise in pioneer technology so developers can focus on solving problems. #techstartup #techVC

ML at a reasonable scale is all about finding a way to solve the right problem. For example, the team over at Continuum Industries needed to figure out a better way of setting up CI/CD for the infrastructure design optimization engine. “We’re still a fairly small team (10 devs or so), so we’d rather avoid having to manage this system ourselves, so we can focus on building our product and improving the AI. We had to do that with our previous system, and it was a huge time sink.” - Andreas Malekos In-house solution < hosted tool (3 main benefits by Andreas): 1# Total flexibility in the metadata structure “We are huge fans of the way data is structured in Neptune runs. The fact that we can basically design our own file structure effortlessly gives us enormous flexibility.” 2# Better focus “Gone are the days of writing stuff down on google docs and trying to remember which run was executed with which parameters and for what reasons. Having everything in Neptune allows us to focus on the results and better algorithms.“ 3# More confidence in the results “The ability to compare runs on the same graph was the killer feature, and being able to monitor production runs was an unexpected win that has proved invaluable.” Pragmatic teams don’t do everything. They focus on what they actually need. Full story: https://buff.ly/3dVnvty #machinelearning

Curious about the future of legacy systems? Explore the world of machine learning with our latest blog by Hassan Abbasi, Software Developer at Authority Partners. 💡 Unlock the secrets of Variational Autoencoders (VAEs) and learn how to enhance system functionality and resilience, as well as how to extract insights from historical data. Let Hassan be your guide in elevating your legacy systems to the next level! 🔗 Read the full blog here: https://lnkd.in/dGgh-qUb #LegacySystems #MachineLearning #TechInnovation #APknowsIT

ML at a reasonable scale is all about finding a way to solve the right problem. For example, the team over at Continuum Industries needed to figure out a better way of setting up CI/CD for the infrastructure design optimization engine. “We’re still a fairly small team (10 devs or so), so we’d rather avoid having to manage this system ourselves, so we can focus on building our product and improving the AI. We had to do that with our previous system, and it was a huge time sink.” - Andreas Malekos In-house solution < hosted tool (3 main benefits by Andreas): 1# Total flexibility in the metadata structure “We are huge fans of the way data is structured in Neptune runs. The fact that we can basically design our own file structure effortlessly gives us enormous flexibility.” 2# Better focus “Gone are the days of writing stuff down on google docs and trying to remember which run was executed with which parameters and for what reasons. Having everything in Neptune allows us to focus on the results and better algorithms.“ 3# More confidence in the results “The ability to compare runs on the same graph was the killer feature, and being able to monitor production runs was an unexpected win that has proved invaluable.” Pragmatic teams don’t do everything. They focus on what they actually need. Full story: https://buff.ly/3dVnvty #ML #MLOps #ReasonableScale

In modern data-driven and machine-learning workflows, efficient orchestration and execution of tasks are crucial in achieving productivity and scalability. To address these needs, we introduced Flyte Agents — long-running, stateless services designed to execute tasks efficiently. 🚀 With agents, you can enhance the overall efficiency and reliability of your data-driven and machine-learning pipelines, ultimately driving higher productivity levels and scaling for your projects. In this blog post, we will provide a step-by-step guide for developing custom agents, exploring real-world use cases where the Flyte Agent emerges as a game changer, unlocking new possibilities for streamlining complex workflows and accelerating data processing tasks. Whether you are a data scientist, ML engineer, or workflow enthusiast, this blog will equip you with the knowledge and tools to harness Flyte Agent's full potential in your projects. 🔗 https://lnkd.in/g63x9dsG

A few months ago, we created a tutorial on how you can develop a financial chatbot using Gradient, LlamaIndex, and MongoDB. With the recent launch of our Gradient Accelerator Blocks, we're showing you how you can recreate this very same financial chatbot 10x faster - without sacrificing quality or performance. ✅ Gradient: Private, state-of-the-art LLMs & embeddings ✅ MongoDB Atlas: Indexing & Vector DB ✅ LlamaIndex: Orchestration & Advanced RAG Framework ✅ Streamlit: User Interface Accelerator Blocks are comprehensive building blocks, designed to help you rapidly build best-in-class AI on a single platform. Test drive all 3 flavors of our accelerator blocks: LLM Development Blocks, Task Specific Blocks, and Domain Specific Blocks. #Gradient #GradientAI #LLM #Finetuning #RAG #FinServ #FinServAI

Speed up your development time with MongoDB Atlas and Gradient Accelerator Blocks. Learn how to build a financial chatbot at a staggering 10x speed and handle storage, indexing, and retrieval of high-dimensional vector data. 📈

A3 provides a framework for navigating the trade-offs in AI application development. A good read for anyone working building LLM Applications.

pub.towardsai.net: The first part of a 2-post series introduces Volga, an open-source, scalable data/feature calculation engine designed for modern real-time AI/ML applications. Volga aims to provide a Pandas-like API for defining data entities, online/offline pipelines, and sources, with consistent online+offline feature calculation semantics. Built on top of Ray, it aims to be a standalone system without heavy dependencies on general data processors or cloud-based platforms, suitable for use on a laptop or a 1000-node cluster. The post also covers the background and challenges of real-time ML, the rise of feature stores and managed feature platforms, and the need for a self-serve feature calculation engine. The second part of the series delves into Volga's architecture and technical details.

The “A3” Paradox: Designing LLM Applications When Less is More  This month I am excited to publish a deep dive into the A3 Theorem, an architectural framework for LLM deployments co-authored by Tapas Moturu, Chief Architect at Intuit, a founding member of Decibel’s AI Center of Excellence. Inspired by the CAP Theorem for distributed data stores, the A3 Theorem details the tradeoffs between Applicability, Adaptability, and Affordability (together, the “A3”), and provides a blueprint to help product & engineering leaders deploy AI models in production. The A3 theorem elegantly demonstrates the paradox of choice when building an LLM application. There are tough choices when scaling language model deployments and it is always better to optimize your product design around one or two “As” to find success. Ultimately the demands of your users will determine the best path - often the speed, accuracy, and breadth of your LLM-native experience will guide your architectural choices and lead to important design choices such as whether to build a synchronous copilot or an asynchronous agent. Once the tough choices have been made there are many ways to attack the problem to create an optimal result. Selecting open vs. closed models, creating LLM Ops data pipelines, and optimizing for elastic GPU infrastructure can move the needle in any AI-native applications that begins to scale. We are grateful to Tapasvi Moturu, Mallik Mahalingam, and the Intuit team for sharing their insights as they have scaled their LLM deployments to over 100 M users. We hope the A3 Theorem serves as a compass and guide for all moving AI pilots to production.