Fully autonomous cars are on the horizon. Knightec wants to enable passengers of the future car to be more productive and entertained with a new web platform. With this platform, Knightec wants to explore different input methods one of which being a voice assistant. A key component in a voice assistant is Automatic Speech Recognition (ASR) and for this task, Knightec had planned to use the new Web Speech API. Their target platform (Oculus Quest 2) does not yet support the Web Speech API and a future implementation could be limited. This thesis conducts a comparative study to find alternatives for running ASR in a web application. The study aimed to compare browser-implemented ASR methods to server implemented methods with Web Speech API as a baseline. The study first conducted a document study to find methods for running ASR tasks inside a web application and then create requirements for method selection. With the requirements, two suitable implementations were found for a browser implementation of ASR. During the final implementation, one of these failed, leaving only one method implemented in the browser. Three ASR methods were chosen for the server implementation, following requirements also set by the document study. To compare the ASR methods a dataset was created with the help of Knightec. The dataset consists of 10 commands, utilizing the voices of six individual employees at Knightec including separate versions, one with and one without background noise for each voice totaling 120 recordings. The dataset was used as a benchmark for each implementation where Word Error Rate (WER) and response time were measured. Due to the structure of the Web Speech API, it was not possible to measure response time for this implementation. The result of the benchmark shows that Web Speech API consistently outperforms the other methods in terms of WER. The response times of the browser implementation could not keep up with the other methods implemented and is not in the range of acceptable results. The recommended implementation for Knightec is to use a server-based implementation while for the general case Web Speech API is the best alternative.

There has been extensive research and progress on virtualization on CPUs for a while. More recently the focus on GPU virtualization has increased as processing power doubles roughly every 2.5 years. Coupled with advances in memory management and the PCIe standard the first hardware assisted virtual solutions became available in the 2010s. Very recently, a new virtualization mode called Multi-Instance GPU (MIG) makes it possible to isolate partitions with memory in hardware rather than just software. This thesis is focused on virtual GPU performance and capabilities for AI training in a multi tenant situation. It explores the technologies currently used for GPU virtualization,including Single Root IO Virtualization (SR-IOV) and mediated devices. It also covers a proposed new standard for IO virtualization called SIOV that addresses some of the limitations in the SR-IOV standard. The limitations of time sliced virtualization are mainly the lack of customization for a partition compared to CPU virtualization and the problem of overhead. MIG virtualization is more customisable in how compute power and memory can be allocated, the biggest limitation is that fast intercommunication is not currently possible between partitions, making MIG more suited for applications that can run on just one partition. It is also not suited for graphical applications as it currently does not support any graphical APIs. The experimental results showed that in compute situations the overhead of time sliced virtualization is around 5% while the maximum intercommunication bandwidth is lowered by 11% and latency increased by 25%. Time slice windows of 4ms compared to 2ms can decrease scheduling overhead to nearly 0.5% at the cost of increased latency for the end user, this can be beneficial for applications where user interactivity is not of importance.

With the increasing integration of artificial intelligence (AI) technologies into educational settings, it becomes important to ensure responsible and effective use of these systems. This thesis addresses two critical challenges within AI-driven educational applications: the effortless integration of different Large Language Models (LLMs) and the mitigation of inappropriate content. An AI assistant chatbot was developed, allowing teachers to design custom chatbots and set rules for them, enhancing students’ learning experiences. Evaluation of LangChain as a framework for LLM integration, alongside various prompt engineering techniques including zero-shot, few-shot, zero-shot chain-of-thought, and prompt chaining, revealed LangChain’s suitability for this task and highlighted prompt chaining as the most effective method for mitigating inappropriate content in this use case. Looking ahead, future research could focus on further exploring prompt engineering capabilities and strategies to ensure uniform learning outcomes for all students, as well as leveraging LangChain to enhance the adaptability and accessibility of educational applications.

I detta examensarbete utforskas vilka överförings- och betalningsmetoder ungavuxna föredrar av Swish, Klarna och Tink samt tekniska implementering gällandekodkomplexitet för dessa överförings- och betalningsmetoder. Detta är viktigt för attkunna erbjuda överförings- och betalningsmetoder som unga vuxna känner sigtrygga med och till att denna målgrupp fortsätter gå mot ett kontantfritt samhälle.Studien syftar till att fylla det forskningsgap som finns kring en direkt jämförelse avdessa betalningsmetoder.Forskningen visar att unga vuxna föredrar metoder som är användarvänliga och harett starkt varumärkesrykte. Forskningen visar också behovet av att vända sigspecifikt till unga vuxna då dessa saknar finansiell kunskap inom området.En prototyp utvecklades för att jämföra kodkomplexitet mellan de mest populärabetalningsmetoderna Swish, Klarna och Tink. Kodkomplexiteten förimplementeringen av dessa metoder mättes med Cyclomatic Complexity, ochresultaten visade att Swish, Klarna och Tink alla har en liknande implementering viaAPI-anrop i sin enklaste form, vilket innebär att valet mellan dem inte innebär någonskillnad i kodkomplexitet.Utifrån jämförandet av kodkomplexitet valdes en komplett prototyp att utvecklas föratt funktionstesta en implementetion av Tink, med hjälp av Quarkus för logik-delenkallad backend och SwiftUI för utseende-delen kallad frontend. SwiftUI valdes föratt skapa en dynamisk och användarvänlig gränssnittsupplevelse för Apple-enheter,medan Quarkus valdes för sin robusthet och effektivitet på server sidan.Prototypen funktionstestades och utvärderades av produktägaren Peak AMSecurities AB. Resultatet blev att prototypens kodbas kan användas i framtidaproduktionssättning.Denna studie bidrar till insikt om unga vuxnas preferenser för digitalabetalningsmetoder och erbjuder en teknisk grund för vidare forskning inomområdet. Genom att jämföra Swish, Klarna och Tink ger arbetet viktiga insikter somkan hjälpa utvecklare att skapa bättre finansiella verktyg som möter användarnasbehov och förväntningar. / This thesis explores which transfer methods young adults prefer among Swish,Klarna, and Tink, as well as the technical implementation concerning code com-plexity for these transfer and payment methods. This is important to offer transferand payment methods that young adults feel comfortable with and to ensure thistarget group continues to move towards a cashless society. The study aims to fill theresearch gap regarding a direct comparison of these payment methods.Research shows that young adults prefer methods that are user-friendly and have astrong brand reputation. The research also highlights the need to specifically ad-dress young adults as they lack financial knowledge in the area.A prototype was developed to compare code complexity with the most popular pay-ment methods Swish, Klarna, and Tink. The code complexity for the implementa-tion of these methods was measured using Cyclomatic Complexity, and the resultsshowed that Swish, Klarna, and Tink all have similar implementations via API callsin their simplest form, which means they do not increase code complexity for devel-opers.Based on the comparison of code complexity, a complete prototype was chosen tobe developed for functional testing, implemented with Tink, using Quarkus for thebackend logic and SwiftUI for the frontend interface. SwiftUI was chosen to createa dynamic and user-friendly interface experience for Apple devices, while Quarkuswas chosen for its robustness and efficiency in server-side processing.The prototype was functionally tested and evaluated by the product owner Peak AMSecurities AB. The result was that the prototype's codebase can be used in futureproduction deployment.This study contributes to the understanding of young adults' preferences for digitalpayment methods and provides a technical foundation for further research in thearea. By comparing Swish, Klarna, and Tink, the work offers important insightsthat can help developers create better financial tools that meet users' needs and ex-pectations.

Laser Speckle Contrast Imaging, or LSCI, is a non-invasive, fast, cheap and easy to use perfusion imaging method which has shown potential in many clinical applications. One problem with the technology however is its limiting field-of-view which results in the physician having to examine and work with multiple images instead of one. In this thesis, a real-time image stitching system was designed with the aim of extending the field of view of LSCI instruments. The system implements a feature-based approach to image registration (SIFT), brute-force matching of features and feather blending. The image transformation is estimated using a statistical methodology (RANSAC) and then validated to improve usability. The evaluation of the system is focused on three key factors: running time of the image stitching algorithm, robustness relative-image pair overlap and usability. The results show that stitching of perfusion images is instantaneous pursuant to human perception for lower resolution images and takes 1.5 to 3.5 seconds for larger resolutions. Stitching is robust given an image-pair overlap of 10% or more, however, temporal noise and sparse environments in perfusion images negatively impacts the systems accuracy and usability. In conclusion, given a contrast-rich region, the system is accurate and easy-to-use such that an untrained user can construct a visually-pleasing panorama that consists of 24 stitched images in approximately 7 minutes.

The Internet plays an important role in modern society, and its network performance impacts billions of users every day. For many network applications, network latency has a large impact on the quality of experience for the end user. Due to a lack of extensive network latency monitoring, the observability of network latency in real networks is often limited. This poses a problem for understanding network latency on the Internet today, and for assessing the impact various solutions that aim to reduce network latency have once they are deployed in the wild. This thesis addresses shortcomings with current solutions for monitoring network latency, in particular the performance of passive monitoring solutions on general-purpose commodity hardware, aiming to enable more ubiquitous latency monitoring and ultimately provide a comprehensive view of real-world network latency. We utilize the recently emerging eBPF technology to implement passive network latency monitoring inside the Linux kernel. Through experiments on a testbed, we show that our solution can monitor packets at over an order of magnitude higher rates than comparable previous solutions, allowing it to successfully monitor the latency for multi-gigabit traffic on general-purpose commodity hardware. Additionally, we demonstrate the feasibility of continuously monitoring network latency by deploying our solution inside an Internet Service Provider and monitoring the network latency for all customer traffic. Through an extensive analysis of the collected latency data, we show large differences in how network latency is distributed across different parts of the network. / The Internet plays a vital role in modern society, and its performance affects billions of users daily. Network latency often has a significant impact on the end users' experience. However, due to limited monitoring of network latency, the observability of latency in real networks is often poor. This hinders our understanding of latency on the Internet today and makes it challenging to assess how the deployment of new networking technologies impacts latency. This thesis uses the emerging eBPF technology to improve the performance of passive network latency monitoring, aiming to enable latency monitoring on more network devices to create a more comprehensive view of latency on the Internet. By conducting controlled experiments on a testbed, we find that our solution is over an order of magnitude faster than previous solutions, making it possible to monitor multi-gigabit traffic on general-purpose commodity hardware. Furthermore, we demonstrate the feasibility of continuously monitoring latency by deploying our solution inside the network of an Internet Service Provider to monitor all their traffic. Our analysis of the latency data reveals large differences in how latency is distributed across different parts of the network.

This project has been created at the growth agency Maia. The goal of the project is to develop satellite sites in WordPress using the Elementor plugin with focus on universal design, SEO (Search Engine Optimization), and accessibility for all user types. The purpose for the satellite sites is to drive traffic to other websites through effective SEO and accessibility. The project work has included planning and reviewing theories around SEO techniques, which has been necessary to ensure that the satellite sites are optimized in the best way to achieve a high search engine ranking. Part of this work has involved developing design sketches that are not only visually appealing but also technically functional to support both SEO and usability. In addition to developing the satellite sites, the project includes the creation of a presentation page developed with Express and React. This page serves as an interactive report on the project with the satellite sites, where specific parts of the work will be shown, as well as results, progress, and examples. / Detta arbete har utförts på tillväxtbyrån Maia. Målet med projektet är att utveckla satellitsidor i WordPress med tillägget Elementor där fokuset ligger på universell utformning, SEO (sökmotoroptimering) och tillgänglighet för alla användartyper. Syftet med satellitsidorna är att med effektiv SEO och tillgänglighet, driva trafik till andra webbsidor. Projektets arbete har inkluderat planering och granskning av teorier kring SEO-tekniker, vilket har varit nödvändigt för att säkerställa att satellitsidorna optimerats på bästa sätt och ge en hög rankning på sökmotorer. En del av detta arbete har inneburit utvecklandet av designskisser som inte bara är visuellt tilltalande utan också tekniskt funktionella med att stöjda både SEO och användarvänlighet. Utöver att utveckla satellitsidorna, innefattar projektet skapandet av en presentationssida utvecklat med Express och React. Denna sida fungerar som en interaktiv rapport över projektet med satellitsidorna där specifika delar av arbetet kommer att visas samt resultat, framsteg och exempel.

This project involves further developing the fitness application Fitness Coach for the company Myrstack AB. The web application is used by trainers to create training schedules and activities that clients can follow via a separate mobile application. The further development of the web application is based on tickets from a product backlog in GitLab. The work focuses on the frontend of the web application, which is developed with Vue3 Composition API, Typescript and Tailwind. During the project, certain functionalities have been added, such as the ability to select weight units (kg/lbs) when creating or editing training activities, as well as the creation of a profile page for trainers. The profile page has been designed with accessibility principles and tested with the WAVE browser tool to ensure it meets accessibility standards. Despite some challenges in understanding and utilizing new technologies, the goals of this project has been achieved. Working with this project has provided deeper insights into version control using GitLab and working with Vue3, as well as various framework extenstions such as VeeValidate and Yup. / Det här projektet handlar om att vidareutveckla träningsapplikationen Fitness Coach för företaget Myrstack AB. Webbapplikationen används av tränare för att skapa träningsscheman och aktiviteter som klienter kan följa via en separat mobilapplikation. Vidareutvecklingen av webbapplikationen baseras på tickets från en produktbacklogg i GitLab. Arbetet fokuserar på webbapplikationens frontend-del som är utvecklad med Vue3 Composition API, Typescript och Tailwind. Under projektet har viss funktionalitet lagts till, som att välja viktenhet (kg/lbs) vid skapande eller redigering av träningsaktiviteter, samt skapandet av en profilsida för tränare. Profilsidan har utformats med tillgänglighetsprinciper och testats med webbläsarverktyget WAVE för att säkerställa att den uppfyller tillgänglighetsstandarder. Trots vissa utmaningar med att förstå nya tekniker och använda dessa så har målen uppnåtts. Arbetet med projektet har givit djupare insikter i versionshantering med GitLab och arbetet med Vue3 samt olika tillägg till ramverket, såsom VeeValidate och Yup.

This project aims to develop a prototype for an analysis tool intended for an upcoming web application within the Central Study Support Authority (CSN). The focus is on designing a user-friendly and accessible tool to analyze the usage of CSN's component library. Through an iterative process, where user experience is prioritized, methods such as user surveys and wireframe creation have been used to clarify the design and structure. The project follows the "Double Diamond" model with phases including Discover, Define, Develop, and Deliver. During these phases, interviews have been carried out to understand user needs and requirements, resulting in a prototype in Figma. After creating the prototype, usability tests were performed to evaluate the tool's effectiveness and usability. The results of these tests were used to optimize the prototype and ensure that the final analysis tool best meets the users' needs. / Syftet med detta projekt är att utveckla en prototyp för ett analysverktyg avsett för en kommande webbapplikation inom myndigheten Centrala studiestödsnämnden (CSN). Fokus ligger på att designa ett användarvänligt och tillgängligt verktyg för att analysera användningen av CSN:s komponentbibliotek. Genom en iterativ process, där användarupplevelsen är i centrum, har metoder som användarundersökningar och skapande av wireframes använts för att konkretisera designen och strukturen. Projektet följer Double Diamond-modellen med faser som Upptäcka, Definiera, Utveckla och Leverera. Under dessa faser har intervjuer genomförts för att förstå användarbehov och krav, vilket resulterat i en prototyp i Figma. Efter att prototypen skapats har användbarhetstester genomförts för att utvärdera verktygets effektivitet och användbarhet. Resultaten av testerna har använts för att optimera prototypen och säkerställa att det slutliga analysverktyget bäst möter användarnas behov.

This master’s thesis explores the application of Artificial Intelligence (AI) in the digitization ofunstructured documents which contains normal text, handwritten text and also integers- a criticalaspect for infrastructure management. As digitization progresses, the efficiency in handling suchdocuments remains a considerable challenge due to their unstructured nature and variedhandwritten quality. The research evaluated several Optical Character Recognition (OCR)models, including Pytesseract, EasyOCR, KerasOCR, docTR, to identify the most effectivemethod for converting handwritten documents into digital, searchable formats. In this study, eachmodel was rigorously tested using a carefully curated dataset containing handwritten and printeddocuments of varying quality and complexity. The models were assessed based on their ability toaccurately recognize characters and words, handle multilingual documents, and process a mix ofhandwritten and printed content. Performance metrics such as Character Error Rate (CER) andWord Error Rate (WER) were used to quantify their accuracy. The results reveal that each model exhibits unique strengths. PyTesseract excelled at convertinghigh-quality images to text with minimal errors, while EasyOCR demonstrated robustrecognition across multiple languages. KerasOCR and docTR proved effective in handlingcomplex, unstructured documents due to their advanced AI architectures. By leveraging thesetechnologies, the thesis proposes an optimized approach that integrates metadata extraction toenhance the organization and searchability of digitized content. The proposed solution,compatible with both CPU and GPU platforms, reduces the time and resources required formanual processing, making it accessible for a broader audience. This research contributes to the field by offering insights into the performance of different OCRmodels and providing a practical, scalable solution for digitizing and managing unstructuredhandwritten documents. The solution promises to significantly improve the efficiency ofdocument management, paving the way for future innovations in this space.