Introducing Julius AI, an intelligent data analyst designed to streamline data analysis and speed up insights discovery.The AI-powered product allows users to explore, interpret, and visualize data using natural language—no coding needed.Julius AI emphasizes ease of use, robust security, and collaborative features, making it suitable for a wide range of professionals from finance to scientific research.In addition, Julius can do data forecasting and statistical analyses, empowering users to uncover patterns, predict trends, and identity anomalies quickly.
Julius supports CSV/XLS, Google Sheets, PDF, text files and can connect directly to various data sources such as PostgreSQL, Google BigQuery, Microsoft OneDrive, Snowflake, Google Drive, GitHub, etc.
The Model Context Protocol (MCP) is an open-source standard created by Anthropic to facilitate better connectivity between AI assistants and various data systems. Historically, AI models have struggled with data isolation, requiring custom integrations for each new data source. MCP aims to solve this issue by providing a universal protocol for secure, two-way data connections, enabling AI systems to access information from content repositories, business tools, and development environments more easily. This allows AI models to produce more relevant responses by understanding the context around a task. Developers can expose data through MCP servers or build AI applications that connect to these servers.
Business Request:Customize data labeling where users can dynamically select what years to display value at line end and what years to display value for every month on line charts.
Solution:We will use set selection and dual axis to customize the data labeling.
1) Create a set for Year of Ship Date that will display values by line end in line chart (@Ship Date To Show Line End Set).
2)Create another set for Year of Ship Date that will display values for every month in line chart (@Ship Date To Show All Values Set).
3)Create formulas to calculate for sales based on each set.
@ Sales (Line End)
If [Ship Date To Show Line End Set] then [Sales] end
@ Sales (All Values)
If [Ship Date To Show All Values Set] then [Sales] end
4)Use these 2 formulas to create a dual-axis line chart.
5)Show sets for users to select years to display value at line end and to select years to display value for every month on line charts.
Many thanks to Jim Dehner for help with the solution.
ServiceNow's Retail Service Management (RSM) platform is a comprehensive solution designed to streamline retail operations and enhance customer experiences. The platform offers a wide array of features including AI-powered workflows, omnichannel support, self-service portals, and workforce optimization tools. It's built on the Now Platform®, a single cloud application platform, and available globally with support in multiple languages. The text details various products, solutions, and resources available within the platform, emphasizing its capabilities in improving efficiency, reducing costs, and boosting customer satisfaction. ServiceNow provides extensive resources, including training and partner support, for successful implementation.
Please visit ServiceNow, the AI platform for business transformation, to explore more about its portfolio of phenomenal products, now supercharged with AI agents.
Business Request:Hide Null value in filter 'Downtime Type' but still keep it in calculation and show only 'Planned' and 'Unplanned'.The reason is having Null value in the filter might be confusing for the users.
Solution:
1) Create a fake field to assign Null a value that exists in the data set for field 'Downtime Type'.
@Fake Field
CASE [Downtime Type]
When null then 'Planned'
Else [Downtime Type]
END
2) From this 'Fake Field', create a set 'Fake Field Set'
3) Create a 'Fake Field Set Filter' to be used as a filter
@Fake Field Set Filter
CASE [Downtime Type]
When null then True
Else [Downtime Type] in [Fake Field Set]
4) Drag @Fake Field Set Filter to filter shelf and select 'True'.Use 'Fake Field Set' as a filter which should have only 2 values 'Planned' and 'Unplanned'.
Many thanks to Diego Martinez for help with the solution.
ServiceNow's Sales and Order Management(SOM) enables organizations to create one continuous value stream by managing the lead to renewal lifecycle on a single platform.SOM helps teams to define new products, services, and catalog-driven fulfillment policies quickly, automates sales and order fulfillment processes, and empowers service staff to manage post-sale activities. The results: Organizations can bring offerings to market faster, accelerate and boost revenue, reduce costs, and improve customer experience.
Please visit ServiceNow, the AI platform for business transformation, to explore more about its portfolio of world-class products, now supercharged with AI agents.
ServiceNow's Manufacturing Commercial Operations platform is a comprehensive solution designed to streamline sales, service, and support operations for manufacturing businesses. The platform offers a suite of applications and capabilities, including AI-powered workflows and tools for order management, customer service, and channel partner engagement, all built on the ServiceNow AI Platform. It provides various packages to suit different business needs and scales with growth, aiming to improve revenue, reduce costs, and enhance customer experiences.
Please visit ServiceNow, the AI platform for business transformation, to explore more about its portfolio of premium products, now supercharged with AI agents.
ServiceNow's Customer Service Management (CSM) enhances every aspect of the customer lifecycle.CSM accelerates self-service resolution capabilities, automates customer operations processes across organization, and empowers agents with real-time intelligence and productivity tools. The platform harnesses the power of the whole organization and leverages AI and automation to improve customer service, reduce costs, and enhance agent productivity.
Please visit ServiceNow, the AI platform for business transformation, to explore more about its portfolio of superlative products, now supercharged with AI agents.
ServiceNow's Workflow Data Fabric is a new foundation designed to unify enterprise data from disparate systems, granting both employees and AI agents real-time, secure, and governed access. This capability enables organizations to connect, understand, and act on any data to power workflows and solve complex business challenges. By eliminating the need to duplicate or transfer data, the platform facilitates faster integration, lower costs, and continuous process improvement. Ultimately, Workflow Data Fabric aims to transform business operations by bringing data, AI, and workflows together on a single platform.
Please visit ServiceNow, the AI platform for business transformation, to explore more about its portfolio of industry-defining products, now supercharged with AI agents.
In quantum computing, a qubit is the fundamental unit of information, analogous to a bit (binary digit) in classical computing, but capable of representing 0, 1, or a superposition of both.
A physical qubit is a real, physical system (like a trapped ion or a superconducting circuit) that can exist in a superposition of states.
A logical qubit is a collection of multiple physical qubits to achieve greater stability and error correction. It's not a physical entity, but a redundant encoding of qubit's quantum state using multiple physical qubits.
NVIDIA GTC 2025 in San Jose has concluded with a big fanfare and what an amazing conference it was.This developers’ conference was the destination for all things AI and more, an incredible place to explore and exchange grand ideas from agentic AI to humanoid robots to quantum computing.
GTC 2025 debuted Quantum Day, showcasing the rapidly progressing field of quantum computing where the impact will be far-reaching.This next-generation field will push the technological boundary with extremely fast computing that's beyond today's fastest supercomputers.Foreseeable future will have QPUs, GPUs, and CPUs working together to extend classical computing.
The challenges of quantum computing are scalability, fidelity, error correction, coherent time, and latency.New methods and algorithms will have to be created in order to apply quantum computing to real-world problems (trillion-dollars business idea here).
Strategic Partnership
ServiceNow expanded NVIDIA partnership to collaborate on Agentic AI capabilities that allow corporate customers to assess AI Agent performance on the powerful ServiceNow Platform.
Please visit ServiceNow, the AI platform for business transformation, to explore more about its portfolio of next-level products, now supercharged with AI agents.
Matryoshka embeddings are an advanced concept in machine learning, particularly in the field of natural language processing (NLP). The term is inspired by the Russian Matryoshka dolls, which are a series of nested dolls that fit inside one another. In the context of embeddings, this nesting idea is extended to create hierarchical relationships between different levels of information. Let’s break down the concept in detail:
What are Embeddings?
Embeddings are vector representations of items like words, sentences, or images, which enable the capture of contextual relationships between those items. These representations allow complex data like text to be used in machine learning models. In language models, word embeddings like Word2Vec, GloVe, or BERT convert words into numerical vectors that capture meaning and semantic relationships between words.
Matryoshka Embeddings - Concept
Matryoshka embeddings leverage the idea of nesting to represent different levels of information within a single structure. This approach enables a hierarchical understanding of data, where the representations capture details at multiple resolutions or granularities.
Hierarchy & Nesting in Matryoshka Embeddings
Nested Hierarchy: Matryoshka embeddings are designed to represent entities at multiple levels of abstraction in a hierarchical manner. For example, in a language model, we could have word-level embeddings, sentence-level embeddings, and document-level embeddings all represented within the same framework.
Shared Information: The embeddings at a higher level (e.g., a document) contain the information of lower levels (e.g., sentences and words) in a nested manner, just like a Matryoshka doll containing smaller dolls. This allows for different granularity of information to be available depending on the specific requirement.
Multi-resolution Understanding: These embeddings allow a multi-resolution understanding of data, meaning you can look at the data in broad strokes (e.g., document-level) or zoom into specific details (e.g., word-level). This flexibility is useful in various machine learning tasks, such as information retrieval, text classification, or semantic similarity.
Technical Implementation
Matryoshka embeddings are often constructed using transformer-based architectures, where attention mechanisms play a key role in allowing the model to learn the hierarchical relationships. Here's how it generally works:
Context Aggregation: Starting with word embeddings, the context for each word is aggregated at the sentence level to create a sentence embedding. This process continues to aggregate context at the paragraph level, and eventually at the document level.
Attention Mechanisms: Multi-head attention layers allow the embeddings to attend to different levels of abstraction, ensuring that information flows from the smallest (word-level) to the largest (document-level) effectively.
Loss Functions for Hierarchy: During training, the loss function is designed to ensure that the embeddings correctly capture information at all levels. For example, they need to preserve relationships at both the local (word-sentence) and global (sentence-document) levels.
Real-World Examples
Document Classification in Legal Domain: Imagine you have a large number of legal documents. Each document is composed of paragraphs, sentences, and words, with different granularities of meaning. Matryoshka embeddings can be used to represent these documents so that you have a representation for individual sentences, paragraphs, and entire documents, allowing you to:
Classify whether a document is related to civil law or criminal law
Extract relevant paragraphs that contain specific clauses
Retrieve similar legal cases based on both sentence-level and document-level semantics
Question Answering Systems: In question answering, hierarchical relationships are crucial to finding accurate answers. For example:
A Matryoshka embedding can represent a book chapter, a paragraph within that chapter, and individual sentences within the paragraph. When answering a user query, the system can use the hierarchical structure to determine which level of information is most relevant, allowing it to zoom into the paragraph or sentence level to extract specific details.
Customer Support Chatbots: A customer support chatbot often has to deal with text at different levels of abstraction. Consider:
A user types a question about a product. The chatbot uses a Matryoshka embedding that nests product-level information, feature-level information, and specific troubleshooting tips.
Depending on how generic or specific the user's question is, the bot can choose the appropriate level of response. For a high-level question like "Tell me about Product X," the system provides a general response, whereas for "How do I reset the password?" it can zoom in on the password reset process.
Video Captioning: Matryoshka embeddings can also be used in video understanding and captioning. Imagine a video is broken down into scenes, shots, and frames:
The embedding for the entire video (document-level) contains aggregated information from scenes (paragraph-level), which are in turn made up of shots (sentence-level), and individual frames (word-level).
This nesting allows for comprehensive video summarization at different resolutions. For example, a short summary can describe the video’s overall theme, while a detailed caption can describe events shot by shot.
Advantages of Matryoshka Embeddings
Scalable Representation: They provide a scalable way to represent complex information, allowing applications to handle both fine-grained and coarse-grained information.
Flexible Context Handling: Depending on the task, different levels of embeddings can be utilized, leading to more efficient and task-specific model performance.
Preserves Context: Matryoshka embeddings ensure that relationships between different parts of data are preserved hierarchically, which is particularly important in tasks involving long documents or structured data.
Challenges
Complex Training: Training Matryoshka embeddings can be computationally intensive, as it involves maintaining coherence across multiple levels of hierarchy.
Attention Overhead: The attention mechanism that facilitates hierarchical representation can add significant overhead, especially for very large documents or highly complex nested structures.
Conclusion
Matryoshka embeddings are powerful for representing hierarchical data, similar to how Russian Matryoshka dolls are nested inside one another. By allowing for multi-resolution, nested representations, they enable deep learning models to handle complex relationships within text, images, or even video data more effectively. They are particularly useful in areas like document classification, question answering, and video captioning, where understanding different levels of context is crucial.
