Filter the articles by selecting the desired tags:
Explaining Text-to-Command Conversational Models
This project is about the explainability and interpretability of Large Language Models (LLMs) to foster trust and ensure their effective use in network management. We have fine-tuned LLMs - text-to-command, that allows users to interact with network APIs through natural language. The model interprets user intent and returns relevant commands. The goal was to explain the model’s reasoning processes, advancing our understanding of how these models make decisions, which is crucial for improving performance, reducing biases, and building more transparent and trustworthy AI systems.
AI Agent for Network Analysis
The project involves leveraging generative AI for various networking tasks. The AI agent allows users to interact with network devices through a chatbot instead of traditional interfaces, enables real-time performance data analysis, suggests root cause analyses and troubleshooting strategies, and provides guidance on configuration tasks with automation capabilities. This AI-driven approach aims to enhance efficiency and accuracy in managing complex network environments.
SD-WAN Insights
Cisco WAN Insights is an innovative product that revolutionizes end-user experience optimization in SD-WAN environments. By introducing proactive path recommendations and visibility workflows, network teams can now proactively troubleshoot their networks and mitigate potential conditions that could impact end-users. The project underwent several iterations, incorporating valuable user feedback to refine its design and functionalities. Boasting cutting-edge technology, WAN Insights offers a multitude of custom and interactive data visualizations that empower users with insights. The innovation of the product and transformative capabilities were recognized with the prestigious 2023 Cloud Computing Product of the Year Award by TMC Cloud Computing Magazine, along with other notable accolades. Its recognition in the industry underscores its significant impact and contribution to the field of network optimization and end-user experience enhancement.
Connectivity Universe
In this project, large quantities of customer data are processed to visualize routers and their connections to the application cloud. Each customer's unique dataset represents a distinct constellation in the universe, allowing users to explore this intricate space in a 3D environment. The tool enables users to search for specific nodes in the network, seamlessly switch between various applications and metrics, and gain quantitative insights. This interactive and dynamic visualization empowers exploration of large datasets in a creative way.
Cognitive Networks
The main focus of the project was to implement cutting-edge technology for cross-layer telemetry, harnessing real user feedback to enhance their quality of experience. This involved employing a sophisticated Machine Learning model to identify complex patterns, perform root cause analysis, and gain deeper insights into user perception. Additionally, a network building mechanism was integrated to optimize the overall quality of experience. To support this observability, predictability, and forecasting technology, the project required a robust and intricate backend, complemented by a user-friendly frontend enriched with interactive animations to guide users seamlessly through the process.
Explaining Gradient Boosted Trees
This innovative visualization tool serves as a powerful aid for users seeking to interpret Gradient Boosted Trees results. By displaying detailed data distribution in each node, showcasing the influence of individual features on predictions, and highlighting the importance of each feature in determining prediction positivity, the tool offers various insights. Notably, its features have led to the filing of a patent, solidifying its uniqueness and potential impact.
Visual Explorer
This is a visualization tool designed to be data-driven and adaptable to various datasets and configurations. Each dataset represents a distinct perspective on a given problem, and loading these datasets triggers the rendering of problem-specific charts while maintaining a consistent interaction workflow. The exploration process involves intricate analysis of time series snippets, expertly represented and tagged in diverse ways to unveil valuable insights. This customizable and automated visual tool empowers users to explore complex data effortlessly and gain deeper understanding with flexibility.
Data Diversity
The project utilizes the Kohonen Self Organizing Map (SOM) algorithm to represent the diversity of customer data. Its primary objective is to convert incoming input patterns of varying dimensions into a discrete map with one or two dimensions through adaptive topological ordering. This approach effectively condenses over 100-dimensional data into a lower-dimensional space while preserving their inherent topological relationships. At the conclusion of the explanation video, an animated visual demonstrates the estimation of probability densities for multidimensional radio data from various customers.
Machine Learning Explaining
A collection of projects has been developed with the primary objective of providing comprehensive explanations and interpretations of machine learning models. These projects focused on visualizing various aspects of the models, such as their training process, prediction outcomes, and decision-making mechanisms. It involved creating a wide array of dashboards, interactive visualizations, and innovative algorithms, all designed to offer valuable insights into the inner workings of these machine learning models. The efforts resulted in the production of informative and engaging videos that further aid in conveying the complex concepts of machine learning in an accessible manner.
Media Quality Labels
This project primarily centered around creating a user-friendly and efficient frontend application to rate media quality. Additionally, the development efforts encompassed building the API backend to support the application's functionalities. Furthermore, an analysis dashboard was crafted to extract meaningful information from thousands of user submissions and provide valuable data insights. The project's key objectives were to streamline the media rating process, ensure seamless communication between frontend and backend components, and empower decision-making through the visualization of user-generated data.
Application Health
The primary emphasis of the project revolved around analyzing geographical data related to internet metrics, including their seasonality and application health. The project's main goal was to identify specific routers, sites, or cities experiencing persistent internet issues and discern patterns of seasonality in these problems. By studying and pinpointing these areas, the project aimed to enhance internet infrastructure and address recurring issues more effectively. The ultimate objective was to improve overall internet performance and user experience in targeted regions.
QoS Explorer
The project featured a straightforward user interface coupled with a sophisticated backend data crunching process, capable of handling various metrics with distinct levels of aggregation and hierarchy. It offered a time exploration feature with value filters and selection drill-down capabilities. The visual components encompassed a range of elements, such as graphs, geographic maps, and matrices, which dynamically updated and animated over time. This combination of intuitive UI and a backend processing allowed users to explore and analyze data in a dynamic and visually engaging manner.
Interactive Visuals in Jupyter Notebooks
The series of projects, aimed to establish a seamless link between cloud-based Jupyter notebooks and interactive visualizations created using popular technologies like React, D3.js, and Three.js. This integration enabled data scientists to efficiently explore their data, conduct in-depth analyses, and implement machine learning techniques, all while witnessing real-time representations through custom and interactive visuals. By combining these powerful tools, data scientists could gain deeper insights and make data-driven decisions with greater speed and precision. The project's focus on user-friendly and interactive data exploration enriched the data science workflow and enhanced the understanding of complex datasets.
Anomaly detection and tunnel failure prediction
The outcomes of this project were conveyed through animated data visuals. The videos showcased how machine learning models leveraged an extensive dataset of 8 billion hours of telemetry data to predict internet tunnel failures and detect anomalous tunnels. Additionally, considerable emphasis was placed on assessing the interpretability and explainability of these models, aiming to enhance their transparency and understanding for stakeholders and end-users.
Metric Circles
This project's main objective is to visualize multi-hierarchical and multivariate data sourced from a vast Spark table comprising millions of records. The key concept involves querying the table and obtaining visualizations directly from the resulting networking metrics. The custom circular representation utilized in the visualizations is optimized to handle and present large amounts of data efficiently and effectively.
NanoSignal
NanoSignal is a signal processing toolbox, developed in C++ with the aim of helping analyze complex and large datasets that nanomotion biosensor outputs. Multiple time-series signals can be imported or appended additionally (single column or labeled multiple column). It has the capabilities to fit and detrend signal, filter it using a moving average, calculate FFT and variance in windowed manner. It plots data in forms of line plots, bar charts and box plots. Multi-threading has been incorporated for the import and analysis of large time-series datasets. A notebook object allows saving graphs, parameters and comments, in order to keep and export valuable information.
Swiss Economy Overview
This project provides an overview of the Swiss economy as of 2016, highlighting key indicators such as GDP growth per capita, government debt, and contributions by various economic sectors. It emphasizes Switzerland’s stable and prosperous economy, driven by a highly developed service sector and specialized manufacturing, especially in high-technology industries. The infographic also details contributions to the economy from different industries and cantons, showcasing the country’s diverse economic strengths.
Clustering Swiss Cities
Deciding to move to a new city or start a business in a different location is challenging and requires understanding the local market. Knowing the preferences and behaviors of people in the area, as well as the presence of competitors, is crucial for success. Location data is an essential asset that can guide these decisions, whether aiming to benefit from competitors' marketing or establish a unique brand presence.
Antimicrobial Resistance Visual
This is an interactive visualization of the antimicrobial resistance evolution from 2001 to 2016 in different European countries. The increasing prevalence of antibiotic resistance among bacterial strains in Europe has become a pressing public health concern. This problem poses significant challenges in effectively treating infections and raises the risk of widespread antibiotic ineffectiveness, impacting patient outcomes and healthcare systems.
Various Graphic Design
This portfolio showcases a diverse collection of graphic design projects, ranging from infographics and logo designs to promotional and marketing materials. The work featured here encompasses projects completed for competitions, companies, academic studies, and personal endeavors.
Scientific Illustrations
This is a non-exhaustive collection of scientifically accurate illustrations, meticulously crafted for a wide range of projects and publications. These illustrations have been instrumental in enhancing the visual communication of complex scientific concepts, aiding in academic publications, research presentations, educational materials, and outreach initiatives.