2025 mai võitjad
Bakalaureuse esimese aasta kategooria:
- TyPy (WWW, PDF) - Kostiantyn Iliushchenko.
- juur.ai (WWW, PDF) - Markus Tamm, Robin Otter.
- Particle Life Simulator (WWW, PDF) - Artjom Geimanen.
Bakalaureuseastme kategooria:
- Model for Saving Energy in Distributed Systems Using QC-LDPC Codes (WWW, PDF) - Karl Kristjan Puusepp.
- Trimmer – A Trim-sheet Workflow Add-on for Blender (WWW, PDF) - Laas Hendrik Lumberg.
- Procedular Generation of Gas Giants (WWW, PDF) - Simon Prii.
Magistriastme kategooria:
- Mastering the Unseen: Approaches to Hard-to-Detect Viral Cytopathic Effect (WWW, PDF) - Aleksandr Makarov.
- IAUNet: Instance-Aware U-Net (WWW, PDF) - Yaroslav Prytula.
- Exploring Social Bias in Language Models through the Lens of Cinema (WWW, PDF) - Liisa Rikanson.
Parim Demo:
Publikupreemia:
2025 mai osalejad
Bakalaureuse esimese aasta projektid:
- B101 TyPy - WWW, PDF
TyPyLang on eksperimentaalne Pythonist lähtuv laienduskeel oma extensiooniga, mis eeltöötleb koodi regulaarsõnades ja AST‑muunduritega, et tuua Pythoni peale range tüübisüsteem (use strict), access_controlled klassid (private/protected/public), readonly muutujad, enum‑struktuurid, interface/implements‑mehhanism, generics (<T>) ning as‑tüübiassertsioonid. Runtime’is rakenduvad.
Kostiantyn Iliushchenko. - B102 Particle Life Simulator - WWW, PDF
A particle simulator written in C++ where particles follow rules to create lifelike behavior.
Artjom Geimanen. - B103 juur.ai - WWW, PDF
Vestlusrakendus, mis võimaldab keelemudeli käest Eesti seaduste kohta küsimusi küsida.
Markus Tamm, Robin Otter. - B104 Ostukorvide võrdlus - WWW, PDF
Veebirakendus ostukorvide haldamiseks ja parima hinna leidmiseks. Lähtekood on saadaval lehel https://github.com/dynamyy/oop.
Egert Heliste, Kevin Markus Korrol, Tõnis Lees. - B105 UTchat - WWW, PDF
UTchat on veebipõhine vestlusrakendus, mis võimaldab Tartu Ülikooli tudengitel omavahel reaalajas suhelda. Lisaks pakub rakendus AI-juturoboti tuge ning kuvab olulist ja kasulikku teavet. Kasutajad saavad hetkega näha, mis päevapakkumisi Delta ja Ülikooli kohvikutes pakutakse ning jälgida reaalajas Tartu ilma. Veebilehel kuvatakse ka jooksvaid uudiseid, mis aitavad kasutajatel aktuaalsete teemadega kursis olla.
Hendrik Jaks, Robin Juul, Karl Markus Kiudma.
Bakalaureuse projektid:
- B01 AudioGuard - WWW, PDF
TerraGuardian Robotics arendab AudioGuardi, tehisintellektil põhinevat veealust seiresüsteemi, mis tuvastab autonoomselt ohte ja kaitseb kriitilist kommunikatsiooni- ja energiataristut.
Paul-Henry Paltmann. - B02 Vectorization Tool for Godot Game Engine - WWW, PDF
VectorizeImage is a custom Godot plugin that turns raster textures into triangulated meshes using random distribution and edge detection without segmentation or with segmentation. Perfect for stylized assets or experimental image-based mesh generation.
Karl Arthur Kade. - B03 Trimmer – A Trim-sheet Workflow Add-on for Blender - WWW, PDF
The trim-sheet workflow is a texturing technique that uses multiple textures (trims) in one image. In the thesis, a new add-on called Trimmer is developed for the Blender software, with the goal of simplifying trim-sheet workflow. Two previous Blender add-ons are analysed, based on which the requirements for Trimmer are established. The completed add-on is published on the Blender extensions webpage. It is compared to the previous solutions and its usability is tested by art students.
Laas Hendrik Lumberg. - B04 Procedular Generation of Gas Giants - WWW, PDF
Lõputöö kirjeldab uut viisi kuidas protseduuriliselt gaasihiiglasi genereerida. Seda tehakse esmaselt tekstuuri lisamisega sfäärile mille järel pannakse tekstuur liikuma, et jäljendada gaasihiiglastel leiduvaid visuaale.
Simon Prii. - B05 AI-Assisted Drawing Using the Stable Diffusion Img2Img Model - WWW, PDF
The goal of this bachelor’s thesis was to develop an interactive demo application that allows users to draw simple sketches, from which AI models detect objects and generate enhanced images in a selected style. The demo is intended for display in the Delta academic building and for presentation at various events. The application’s user interface was built using the Gradio framework, and image processing was performed using the Florence-2-large, Stable Diffusion v1.5, and SD ControlNet – Scribble models.
Rainer Kõiv. - B06 Labour and Duty – a VR Production Line Rhythm Game - WWW, PDF
Labour and Duty is an assembly line themed rhythm game for Meta Quest virtual reality headsets, where the player assembles various common products while doing every required step on time to the beat. There are very few different types of rhythm games for virtual reality devices and Labour and Duty is intended to be a novel experience for virtual reality gamers while avoiding virtual reality related discomfort and cyber sickness.
Joosep Tamm. - B07 Automated Podcast Generator - WWW, PDF
Automated podcast generator aims to offer students a new way of learning by allowing them to upload their lecture slides (PDF format) and convert them into podcasts for a more enjoyable studying process.
Philip Paškov, Oskar Männik. - B08 Game Tour – a Visual Development Environment for a Learning Game - WWW, PDF
Game Tour is a no-code development environment created to support the expansion of the educational game Virtual Tour Game. It allows users to visually add and configure interactive elements such as navigation buttons, video displays, and characters directly within the game world, without any programming knowledge.
Mihkel Kulu. - B09 DeltaVR – User Flow Improvements - WWW, PDF
This project discovered, addressed and solved the user flow issues of the VR application DeltaVR.
Ranno Samuel Adson. - B10 Tartu Ülikooli arvutimuuseumi interaktiivsed eksponaadid - WWW, PDF
Projekt hõlmab endas TÜ arvutimuuseumi jaoks loodud kolme interaktiivset eksponaati, mis tutvustavad magnetiliste andmekandjate kasutust.
Artur Reitalu. - B11 Visualization of Disease Frequencies on OMOP CDM Database - WWW, PDF
Processing health data paves the way to improve public health, the development of new treatment methods, and the planning of preventive and development activities. Existing solutions for visualizing health data focus on specific data types and do not provide harmonized visualization capabilities. Within the health-data domain, the Observational Medical Outcomes Partnership Common Data Model (OMOP CDM) is becoming an increasingly adopted data model, defining both the data-model format and standardized terminologies. The goal of this bachelor’s thesis was to create an interactive data dashboard that can be applied to any database in OMOP CDM form. The completed dashboard makes it possible to examine disease frequencies and their associated costs by diagnosis and demographic filters. In addition, the thesis describes a reusable data query that counts the number of unique patients and the total cost of healthcare services provided to them by calendar year, age, and gender group.
Rasmus Mirma. - B12 Box Constructor - WWW, PDF
Greyboxing tool for the Godot Game Engine.
Hannogert Otti. - B13 Terraparsec - WWW, PDF
A 2D Simulation Game of Exoplanetary Survey and Space Exploration.
Rasmus Peäske. - B14 Generatiivne õppeainete soovitaja - WWW, PDF
Generatiivne õppeainete soovitaja on ÕIS II andmetel baseeruv soovitussüsteem, mille eesmärgiks on lihtsustada uute õppeainete avastamist Tartu Ülikoolis.
Rannar Zirk. - B15 AI-Powered Food Safari: Restaurant Picks For The Modern Traveller - WWW, PDF
This project introduces an AI-powered restaurant recommendation system that uses natural language processing and aspect-based sentiment analysis. The system evaluates restaurants based on customer reviews, location, user preferences, and metadata to generate a personalized, and context-aware list of top restaurant suggestions.
Anton Berik. - B16 Model for Saving Energy in Distributed Systems Using QC-LDPC Codes - WWW, PDF
An energy-aware storage architecture based on a quasi-cyclic LDPC code and a novel single-bit decoding method combining Gaussian elimination and peeling decoding.
Karl Kristjan Puusepp. - B17 Fotorealistliku Kuu pinnase materjali protseduuriline genereerimine - WWW, PDF
Lõputöö raames loodi Kuu pinnast võimalikult fotorealistlikult jäljendav materjal, ehk teisi sõnu 3D stseeni objekti välimust määrav reeglistik. Materjal kasutab ainult protseduurilist genereerimist, mis väldib üldlevinud tekstuuripõhiste lahenduste puudusi ning pakub kasutajale rohkem parameetreid välimuse kohandamiseks.
Oliver Jõgar. - B18 Reflexions - WWW, PDF
This project has 6 different interactive digital artworks inspired by nature. The users can interact with all the visuals in real-time by moving in front of a depth camera.
Kreete Kuusk, Henri Maandi. - B19 Kinnisvara kaardirakendus & hindade ennustusmudel - WWW, PDF
Tallinna, Tartu ja Pärnu kinnisvaraturu visualiseerimine ning hindade ennustamine aadressi ja muude parameetrite abil.
Remo Tsernant.
Magistri projektid:
- M01 FÄRG: Eesti keelele optimeeritud klaviatuur - WWW, PDF
Klaviatuuripaigutus, mis on loodud mugavaks kirjutamiseks, võttes arvesse eesti keele sagedasemaid tähti.
Anu Kadri Uustalu, Pirje Õunpuu. - M02 CoralChain – A Visual Gadget Collection for Exploring Blockchain Fundamentals - WWW, PDF
CoralChain is a lightweight, web-based blockchain simulation framework built with Ruby and Sinatra. It brings together a collection of visual, interactive gadgets designed to illustrate key concepts such as blockchain structure, multi-node synchronization, and comparison of consensus mechanisms. This project offers an approachable and hands-on way to explore fundamental blockchain behaviors through simplified, illustrative scenarios.
Xinjian Zhang. - M03 nullis - WWW, PDF
Nullis is a tile-based puzzle game with two different symbols - 0 and 1. To win you must destroy all zeroes.
Veronika Kukk. - M04 Modeling the Spatial Determinants of Urban Air Pollutants: A Multi-Method Analysis of CO₂, PM2.5, NOx, and SO₂ in Functional Urban Areas - WWW, PDF
This project investigates the spatial drivers and propagation mechanisms of urban air pollutant emissions across 619 Functional Urban Areas (FUAs) in Europe. We focus on four major pollutants—CO₂, NOx, SO₂, and PM₂.₅—and examine their relationships with 14 socioeconomic and spatial variables. The study integrates machine learning models with SHAP-based interpretability, geographically weighted regression (GWR) to capture spatial heterogeneity, and graph attention networks (GAT) to model inter-FUA influence. This multi-method approach enables a comprehensive understanding of both local emission patterns and spatial interaction effects across FUAs.
Zekang Chen, Jiahao Zhu. - M05 DeltaVR: Digital Twin - WWW, PDF
This thesis presents a virtual reality-based digital twin of the Delta Centre building. The platform integrates real-time data visualization through interactive dashboards. The project outlines potential features and experiences that could enhance user engagement and functionality. Additionally, it explores various methods for presenting and testing the application to different audiences.
Muhammad Taimour Khan. - M06 Exploring Social Bias in Language Models through the Lens of Cinema - WWW, PDF
To address the limitations of manual dataset creation for bias measurement, we develop a scalable data curation framework using Cinema subtitle data. The resulting dataset, CineBias, contains 1012 sentence pairs reflecting common stereotypes. Using CPS, we measure biases in several language models and demonstrate the method’s cross-linguistic applicability with an Estonian case study.
Liisa Rikanson. - M07 Mastering the Unseen: Approaches to Hard-to-Detect Viral Cytopathic Effect - WWW, PDF
While past global pandemic highlights the acute need, the continuous effort to understand and combat diverse viral infections requires precise and efficient laboratory tools. Viral Cytopathic Effect (CPE), the visible alterations in cells due to viral activity, is a backbone for a wide array of common virological studies. However, manual microscopy for its detection is slow, subjective and struggles with subtle (“hard-to-detect”) infections. Goal: to develop automated computer vision pipeline for the rapid, objective, and quantitative analysis of CPE. We present our work focused on the xenotropic murine leukemia virus (x-MuLV) to demonstrate this capability.
Aleksandr Makarov. - M08 IAUNet: Instance-Aware U-Net - WWW, PDF
Instance segmentation is critical in biomedical imaging to accurately distinguish individual objects like cells, which often overlap and vary in size. Recent query-based methods, where object queries guide segmentation, have shown strong performance. While U-Net has been a go-to architecture in medical image segmentation, its potential in query-based approaches remains largely unexplored. In this work, we present IAUNet, a novel query-based U-Net architecture. The core design features a full U-Net architecture, enhanced by a novel lightweight convolutional Pixel decoder, making the model more efficient and reducing the number of parameters. Additionally, we propose a Transformer decoder that refines object-specific features across multiple scales. Finally, we introduce the 2025 Revvity Full Cell Segmentation Dataset, a unique resource with detailed annotations of overlapping cell cytoplasm in brightfield images, setting a new benchmark for biomedical instance segmentation. Experiments on multiple public datasets and our own show that IAUNet outperforms most state-of-the-art fully convolutional, transformer-based, and query-based models and cell segmentation-specific models, setting a strong baseline for cell instance segmentation tasks.
Yaroslav Prytula. - M09 Cell Cycle Phase Classification from Microscopy Images - WWW, PDF
Accurate classification of cell cycle phases is essential for cancer research and drug discovery. While fluorescence (FL) microscopy provides reliable results, it requires staining procedures that are costly, time-consuming, and harmful to cell viability. Bright-field (BF) microscopy offers a label-free alternative but poses challenges due to lower contrast. We develop five computational strategies to classify cell cycle phases from both fluorescence and bright-field microscopy images using classical machine learning and deep learning models.
Ali Zeynalli.