DEVnet’s Post

One week has passed after the Schwarz IT Bulgaria's DataCon Sofia and I wanted to spend some time to recap the great topics that were discussed. On Day 1, we talked about the importance of the alignment of data strategy to business strategy and how it gives confidence and direction to the technology functions in an organisation. The 3 pillars that need to be present for a successful data strategy - data governance, data quality and data catalogues were mentioned (all very important from my humble experience to create trusted data products that the business can use to achieve its goals with data driven decisions and experiments). Related to that, Ab Initio Software showed ability to scale up the use of that via its self-service BI features in Easy>data and Snowflake introduced a whole host of Machine Learning (ML) focused features that can be useful for both business users and ML experts that really expands their already large capabilities and use cases. On the second day, we talked also about the hot topic of GenAI, the need (or not) of regulation in the space and how that would hamper/influence innovation in Europe. Overall, a great series of talks and discussions, thank you Pencho Dobrev and Daniela Petrova (and your whole team that was responsible for organising it) 

The beauty of DataStax Enterprise 6.9 / HCDP release (preview release out) coming out in mid-July is it runs on Kubernetes supporting integration with built in OpenSearch and Apache Pulsar. This means a direct connection for Kafka via Pulsar connector instead of building custom service code to connect. But the biggest selling point for developers is the Data API which allows you to not have to write CQL instead Mongo like queries using the API interface over REST. This is a gamechanger for DataStax Enterprise companies that run their clusters on prem. Lastly, Vector DB support is also supported in 6.9 which allows for vector data storage directly in DataStax Enterprise. Vector DBs store the data that can be used to seed your LLMs for Gen AI. Lots of amazing product updates coming from DataStax this year. If any of this is exciting to you, love to connect with your team and discuss how we can help your team achieve and accomplish more this year! https://lnkd.in/gMM8uZCh

What Is RAML? RAML, which stands for "RESTful API Modeling Language," is a YAML-based language for describing RESTful APIs. It is a blueprint before developing an API that makes it easier for both developers and stakeholders to understand the API's structure and behavior. What Are Data Types? https://lnkd.in/eTpWTFuW

✨𝐇𝐨𝐰 𝐭𝐨 𝐜𝐡𝐨𝐨𝐬𝐞 𝐛𝐞𝐬𝐭 𝐕𝐞𝐜𝐭𝐨𝐫 𝐃𝐚𝐭𝐚𝐛𝐚𝐬𝐞 : ✨ Consider following key points while comparing different vector databases and choosing the best one : 1.𝐎𝐩𝐞𝐧 𝐒𝐨𝐮𝐫𝐜𝐞 𝐯𝐬. 𝐇𝐨𝐬𝐭𝐞𝐝 𝐒𝐨𝐥𝐮𝐭𝐢𝐨𝐧: Open source software is free and customizable but may lack professional support, while hosted solutions are managed by vendors, offering convenience but potentially higher costs and less control. 2.𝐏𝐫𝐢𝐜𝐢𝐧𝐠: Look at the cost models, such as pay-as-you-go, tiered subscriptions, or custom enterprise pricing, based on your data volume and usage. 3.𝐒𝐮𝐩𝐩𝐨𝐫𝐭𝐞𝐝 𝐕𝐞𝐜𝐭𝐨𝐫 𝐋𝐞𝐧𝐠𝐭𝐡𝐬: Check what types and lengths of vectors the database supports, which affects how complex your data can be. Supported Distances: Determine the similarity metrics available (e.g., Euclidean distance, inner product) for comparing vectors. 4.𝐍𝐞𝐚𝐫𝐞𝐬𝐭 𝐍𝐞𝐢𝐠𝐡𝐛𝐨𝐫 𝐒𝐞𝐚𝐫𝐜𝐡: Consider the methods for finding similar vectors, whether exact (more precise but slower) or approximate (faster but less accurate). 5.𝐂𝐥𝐮𝐬𝐭𝐞𝐫𝐢𝐧𝐠: Look for built-in clustering features for grouping similar vectors, useful for data analysis. 6.𝐅𝐢𝐥𝐭𝐞𝐫𝐢𝐧𝐠 & 𝐀𝐠𝐠𝐫𝐞𝐠𝐚𝐭𝐢𝐨𝐧: Evaluate the database’s ability to filter data and perform aggregation operations to refine search results and summarize data. Integrations: Ensure compatibility with your programming languages and tools through available SDKs and APIs. 7.𝐂𝐥𝐨𝐮𝐝 𝐏𝐫𝐨𝐯𝐢𝐝𝐞𝐫𝐬: Consider the cloud platforms where the database can be deployed and the implications for vendor lock-in. 8.𝐃𝐞𝐯𝐞𝐥𝐨𝐩𝐞𝐫 𝐄𝐱𝐩𝐞𝐫𝐢𝐞𝐧𝐜𝐞: Assess the ease of use, documentation, and community support to ensure a smooth development process. Refer following chart to choose best vector database . Follow for more such updates in AI ! #AI #LLMs #RAG #LanguageModels #DeepLearning #MachineLearning #Langchain #Huggingface

A key lesson I’ve learned, both from experience and the well-known paper 𝘏𝘪𝘥𝘥𝘦𝘯 𝘛𝘦𝘤𝘩𝘯𝘪𝘤𝘢𝘭 𝘋𝘦𝘣𝘵 𝘪𝘯 𝘔𝘢𝘤𝘩𝘪𝘯𝘦 𝘓𝘦𝘢𝘳𝘯𝘪𝘯𝘨 𝘚𝘺𝘴𝘵𝘦𝘮𝘴, is that when we talk about deploying machine learning models, the actual ML code is just a tiny part of the overall system. In fact, it usually accounts for less than 5-10% of the overall effort. In software development, technical debt refers to the cost of future work caused by shortcuts taken during development. This "debt" accumulates when faster, easier solutions are chosen, potentially leading to more maintenance later. Like financial debt, technical debt grows over time if ignored, leading to long-term challenges that become difficult to reverse. Here’s how this concept plays out in ML systems: 𝐃𝐚𝐭𝐚 𝐏𝐢𝐩𝐞𝐥𝐢𝐧𝐞𝐬: Building and maintaining data pipelines takes significant time. Data doesn’t come clean and ready for models. What starts as a simple pipeline can quickly become a “pipeline jungle,” where changes are risky and error-prone. Regular audits and simplifications help reduce complexity. Tracking how long it takes to process new data or make changes can measure technical debt. 𝐆𝐥𝐮𝐞 𝐂𝐨𝐝𝐞: In ML systems, "glue code" is essential. It integrates the model with the broader system, managing data inputs and predictions. Over time, this code can make it harder to upgrade or change components. Standardizing APIs and automating components help reduce the need for heavy refactoring later. 𝐌𝐨𝐧𝐢𝐭𝐨𝐫𝐢𝐧𝐠 & 𝐅𝐞𝐞𝐝𝐛𝐚𝐜𝐤 𝐋𝐨𝐨𝐩𝐬: After deployment, a model requires ongoing monitoring, especially in live systems. Feedback loops, where predictions influence future data, can cause unpredictable behavior or degraded performance. Real-time monitoring and tracking data shifts help prevent the accumulation of hidden technical debt. 𝐂𝐨𝐧𝐟𝐢𝐠𝐮𝐫𝐚𝐭𝐢𝐨𝐧 𝐎𝐯𝐞𝐫𝐥𝐨𝐚𝐝: ML systems involve many configurations—features, parameters, data sources—and it’s easy to lose track. Misconfigurations can lead to failures or inefficiencies. Clear versioning practices and regular configuration reviews help prevent system crashes or slowdowns due to outdated settings. 𝐄𝐧𝐭𝐚𝐧𝐠𝐥𝐞𝐦𝐞𝐧𝐭: One of the biggest challenges in ML systems is entanglement—where changing one feature or parameter affects the entire system. This CACE principle (Changing Anything Changes Everything). Isolating models and serving them as ensembles can help contain changes. Monitoring shifts in prediction behavior early can prevent major disruptions. The takeaway is clear: building the model is just the start. Without attention to these details, systems become fragile, expensive to maintain, and difficult to improve. These under-discussed factors are crucial for building scalable, adaptable systems that can evolve. #MachineLearning #MLOps #TechnicalDebt #DataEngineering #MLSystems

🚀 Exciting news from DataStax! Here is what we know: 👉RAGStack, an out-of-the-box retrieval augmented generation (RAG) solution, is now powered by LlamaIndex, offering developers a streamlined journey to building generative AI applications. This comprehensive solution addresses the challenges faced by enterprise developers, providing curated Python distribution for seamless integration with Astra DB, DataStax Enterprise (DSE), and Apache Cassandra®. 👉Key benefits of RAGStack with LlamaIndex include enhanced indexing and parsing capabilities, enabling clean extraction of tables from complex PDFs with LlamaParse. This partnership with LlamaIndex signifies a significant milestone in simplifying RAG implementation and offering long-term support for enterprise developers. 👉Jerry Liu, co-founder and CEO of LlamaIndex, expressed excitement about reshaping the RAG landscape and offering a simplified journey for enterprises and developers alike. This collaboration embodies a commitment to simplicity, innovation, and limitless possibilities in GenAI applications. 👉Jeff Schneider, CEO at Imprompt, highlighted the importance of RAGStack in enhancing enterprise offerings while prioritizing privacy. The integration with RAGStack enables secure data exchange, powering innovation for global enterprises. 👉Davor Bonaci, CTO and executive vice president at DataStax, emphasized the significance of out-of-the-box RAG solutions in meeting the high demand from enterprise developers. By incorporating LlamaIndex into RAGStack, DataStax is providing developers with a comprehensive GenAI stack that simplifies RAG implementation. 👉Join the live "how to" event on March 12 at 10 a.m. PT with LlamaIndex and Imprompt to learn more about the integration and its benefits. Sign up now for details and get started with notebook examples showcasing how to use LlamaIndex and LlamaParse with Astra DB. What are your thoughts on the integration between DataStax RAGStack and LlamaIndex? Comment below 👇 #RAGStack #GenAI #LlamaIndex #DataStax #Innovation

📢 New tech note available📝 Read the comprehensive tech note authored by Estafet's consultant Delcho Delov on "Collection, processing, and forwarding of log data from various sources using Fluent Bit and Fluentd". 💻 In today's data-driven landscape, efficient log management is paramount, and this tech note dives deep into leveraging Fluent Bit and Fluentd for seamless log data handling. 💡 Delcho Delov's expertise shines through as he intricately explores the nuances of collecting, processing, and forwarding log data from diverse sources, offering invaluable insights and best practices along the way. 📚 Embrace the power of Fluent Bit and Fluentd to optimize your log management processes and unlock new potentials in data analytics. Check out the tech note here https://lnkd.in/dQviBP2x and let's embark on this journey of enhanced log data management together! 🌟 #LogManagement #DataAnalytics #FluentBit #Fluentd #TechInsights #Innovation #DataManagement

Innovation and efficiency are calling! Join us to learn how function calling in #Gemini gives #developers structured outputs and full control when connecting #LLMs to real-time data in external systems via #APIs.

Innovation and efficiency are calling! Join us to learn how function calling in #Gemini gives #developers structured outputs and full control when connecting #LLMs to real-time data in external systems via #APIs.

Innovation and efficiency are calling! Join us to learn how function calling in #Gemini gives #developers structured outputs and full control when connecting #LLMs to real-time data in external systems via #APIs.