Our projects

A multi-semester collaboration with TV 2: machine learning on real article data to understand what engages readers – from engagement prediction to AI-optimized headlines.

Alice Zheng · Afras Mansoor · Olivia Tan · Håvard Fossdal · Folke Jernbert · Kristoffer Nohr Olaisen · Maia Austigard · Sanne Jamila Olsen · Serina Serife Erzengin

One reinforcement-learning project across several semesters: AlphaZero self-play, MuZero planning, TrackMania racing and a shared game-agent arena. Explore the implementations and teams below.

UGO is a No-Limit Texas Hold'em agent trained through reinforcement learning and self-play against earlier versions of itself.

Simone Elia · Federico Hrvatin · Even Ytterli Tokle · Miheer Gautam · Daniel Meling

A deep learning project that segments brain tumors in MRI scans, distinguishing tumor core, edema and enhancing regions.

Eldar Alvik · Muhammad H. Muwaffaq · Frida Kroken · Tobias Amundrud · Sofia Kamali

Exercise recognition, repetition counting and form feedback using IMU sensors and pose estimation.

Dennis Jevne · Vittorio Avellone · Jens Valderhaug · Gustav Natvig · Ivan Kochura · Henrik Øen

An agentic AI system that predicts stock movements and makes recommendations – historical data, feature engineering and gradient boosting in practice.

Rasmus Jonssønn Skramstad · Anders Nikolai Holsen · Vegard Aa Albretsen · Aleksandar Mirkovic · Kristoffer Eide

A deep learning model estimates coordinates from street-view images using visual features such as vegetation, road signs and architecture.

Daniel Neukirch Hansen · Jens Martin Norheim Berget · Magnus Bryne · Sondre Pettersen · Per Henrik Bergene Holm · Parleen Brar · Romeo Henriksen · Håkon Støren · Vetle Støren

The EU AI Act is hundreds of pages of legal text – RegulAId is the agentic assistant that navigates them for you. LLMs and agent technology applied to a problem the entire industry is wrestling with.

Serina Erzengin · Inês Trigueiro · Rasmus Skramstad · Vegard Aa Albretsen · Ingunn Tonetta Erdal

Vision models (BLIP-2) and LLMs in perfect harmony: upload a photo, get a tailor-made pickup line. The science of flirting, served by a vision transformer.

Jozef Kraus · Kristian Carlenius · Christopher

Cogito's official team in the Norwegian AI Championship: computer vision, reinforcement learning and MLOps under competition pressure – and Cogito members took both 1st and 2nd place.

Brage Kvamme · Sverre Nystad · Ludvig Øvrevik · Tobias Fremming · Afras Mansoor · Thomas Sørensen · Serina Serife Erzengin · Mina Al-Dolaimi · Maia Austigard · Jessica Liu

A robotic hand that obeys your muscles: EMG sensors read the signals from your forearm, the AI interprets your intention, and the hand grips. Mechanical design, embedded systems and machine learning in one project.

Tobias Fremming · Afras Mansoor · Even Ytterli Tokle

Cogito's own robot platform across several generations – firmware, hardware and autonomy built and rebuilt by new cohorts of members.

Christian Gynnild · Dennis Jevne

Cogito's exploration of audio and machine learning – signal processing and deep networks meet music.

Sharuca Elangarajah · Muhammad H. Muwaffaq · Nathania Muliawan · Sverre Nystad

A zbot walking simulation using reinforcement learning in JAX with GPU training.

Sebastian

The intro group that gives new members a taste of everything: from Git and ML fundamentals via Flappy Bird RL and an LLM chatbot to tumor segmentation on a supercomputer – all in a single semester.

Sveinung Myhre · Maia Austigard · Mina Al-Dolaimi · Kristian Flatval

A multi-semester collaboration with Infor: cross-sell recommendations, dynamic pricing with EA Smith, and AI agents that turn test instructions into repeatable browser tests. Explore each edition below.

A chess assistant for AR glasses that recognizes a physical board and displays suggested moves in the player's field of view.

Simon Sandvik Lee · Vetle Støren · Håkon Støren · Alice Zheng · Jessica Liu

Real-time motion recognition and interaction, built in C# – your body as the controller.

Tobias Fremming · Maia Austigard · Parleen Brar

A small Python command-line tool that combines Norwegian words, numbers and special characters into passwords. Users can configure word counts, capitalization and the number of passwords to generate.

NEAT is neat! Neuroevolution that doesn't just train the network but breeds the architecture itself – and uses it to play Super Mario Bros.

Christian Fredrik Johnsen · Brage Kvamme · Kristian Carlenius · Ludvig Øvrevik · Kacper Pawlowski · Vetle · Håkon

Digital creatures learning to walk: agents in a 2D environment evolve body and brain through evolution, one stumbling generation at a time.

Jonas Korkosh · Mina Al-Dolaimi · Nathania Muliawan · Nils Henrik Hoelfeldt Lund · Parleen Brar · Sindre Fossdal · Skage Reistad · Tobias Fremming

From chaos to solved cube: CNNs read the cube, and a combination of graph search, Domino Reduction and classic solving algorithms finds the way home.

Eduard Prokhorikhin · Einride Osland · Even Ytterli Tokle · Gustav Natvig · Kien Le

Just A Rather Very Intelligent System: a voice-controlled AI assistant inspired by Iron Man, autonomously solving complex tasks in the background while you talk to it naturally.

Eldar Alvik · Igor Iwanicki · Jon Bergland · Konrad Seime · William M. Schmidt · Ivan Mordovets · Nils Henrik Hoelfeldt Lund · Erlend Skaarberg · Muhammad H. Muwaffaq

A content factory on autopilot: AI generates TikTok videos and publishes them itself – the entire pipeline from idea to uploaded video, with no human in the loop.

Henrik Kvamme · Andreas V. Jonsterhaug · Håvard Daleng

Tetris agents use genetic algorithms to evolve heuristics for placing pieces.

Eduard Prokhorikhin · Henrik Haaland · Håvard Fossdal · Jon Bergland · Maia Austigard · Øystein Kvandal · Sindre Fossdal

Computer vision reads a Sudoku board from a camera image. A search algorithm solves the puzzle and overlays the solution on the image.

Andreas V. Jonsterhaug · Igor Iwanicki · Afras Mansoor · Lotta Helen Heed Jensen

A diffusion model implemented from research papers and trained to generate car images by gradually removing noise.

Marijan Soric · Thomas Haslund Wik · Mauritz Skogøy · Amanda Truyen · Baris Batur

A personal fitness coach you talk to: the LLM agent logs your workouts, analyzes your progress and builds plans toward your goals.

William M. Schmidt · Tryggve Klevstul-Jensen

Upload an entire textbook and get a personal tutor that knows the syllabus – and cites it. RAG, OCR and function calling in the full-stack app that is Cogito's most-starred repo on GitHub.

Henrik Halvorsen Kvamme · Kaamya Shinde · Kristoffer Nohr Olaisen · Olav Selnes Lorentzen · Parleen Brar · Simon Sandvik Lee · Skage Reistad · Sverre Nystad · Tobias Fremming

Type in a topic, get a finished poster or meme – image and text generated in one go. Prompt engineering, finetuning and image generation put into a system, long before everyone else did it.

Daniel Neukirch Hansen · Eduard Prokhorikhin · Håvard Fossdal · Jørgen Haugdal Jore · Kristoffer Nohr Olaisen · Olav Selnes Lorentzen · Sveinung Myhre · Sverre Nystad · Simon Sandvik Lee

Injury prediction for Rosenborg Ballklub: GPS data from the players' training sessions and matches, analyzed with machine learning to find the patterns that lead to injuries – before they happen.

Erik Nymo Bohne · August Lind · Viktor Grevskott · Lasse Lemka · Daniel Bao Duy Nguyen · Joel Constantions

Recommender systems meet CNNs: swipe through cats and let the model learn exactly what kind of cat is your perfect match.

Ulrik Røsby · Marijan Soric · Adi Singh

Give GPT-2 the beginning of a story and let it continue the tale. This desktop prototype pairs a simple text prompt with language-model sampling, displaying the generated story in a Tkinter interface.

Medical image analysis with U-Net: segmentation and classification of real health data using convolutional neural networks – the architecture that revolutionized medical computer vision.

Vilhjalmur Arnar Vilhjalmsson · Sveinung Myhre · Erik Nymo Bohne · Ine Zhao

An NLP model that reads your text and suggests the emojis to match – language understanding in miniature. 🎯

Ulrik Røsby · Øystein Kvandal · Fredrik Marthin Hellevang

A world of artificial life and death: agents are born, compete, adapt and die in a simulated ecosystem.

Birk Stoveland · Jonathan Jørgensen · Øystein Kvandal · Albert Bayazidi

Generative art that breathes: a painting that lives, moves and never shows the same image twice.

Didrik Lindløv Sand · Espen Boman Fosseide · Ulrik Røsby · Hans Jakob Håland · Ann-Marie Revillard · Magnus Rasmussen · Martin Halleland

Deep Q-learning teaching itself Flappy Bird – pixels in, flapping out, one frustrating pipe gap at a time.

Sanne Jamila Olsen

From pixelated to sharp: deep learning that upscales images and reconstructs details that no longer exist in the original.

Jon Rodtang · Espen Boman Fosseide · Sindre Saastad · Ingeborg Malene Klemsdal · Julie Lervaag Breivik

AI that composes music – neural networks learning harmony and melody, creating entirely new pieces.

Theodor Forgaard · Espen Boman Fosseide · Ulrik Isdahl · Mats Ellingsen · Sanne Jamila Olsen

Experiments in finding faces and comparing facial encodings across photographs. The project includes face-location, face-matching and multi-person examples, plus a digital makeup experiment using facial landmarks.

A tournament environment for AI agents inspired by Pommerman: build your bot, drop it into the arena and see if it survives the bombs.

Jonathan Jørgensen

An agent learns to build momentum and drive up a steep hill in Gym's MountainCar environment. Tabular Q-learning turns discretized position and velocity into actions, with training rewards plotted over time.

Faces that never existed: the GAN architecture behind thispersondoesnotexist, explored and trained by Cogito members.

Tobias Christiani · Andreas Gravrok · Espen Boman Fosseide · Henrik Øberg Myhre

Learning to balance a pole on a moving cart: a NumPy Q-table, exploration and reward updates form this early Cogito reinforcement-learning implementation, with a running reward curve to follow training.

Fake news classification with machine learning – NLP applied to one of the internet's biggest problems, long before LLMs took over.

Espen Boman Fosseide · Birk Stoveland · Einar Sommerfeldt

Cogito's first encounter with the DeepMind classic: AlphaZero built up from self-play and Monte Carlo tree search, trained to master Tic-Tac-Toe and Connect Four – with zero human knowledge of the games.

Agnar Martin Bjørnstad · Andreas Gravrok · Ole Fredrik Borgundvåg Berg · Sebastian Aas · Tor Dyrdal Strande · Dilawar Mahmood

Conversational AI before ChatGPT existed: natural language understanding built with DialogFlow and Kotlin.

Simen Burud · Jakob Madsen · Silas Eic · Marius Torbjørnsen

What will you look like in 40 years? Deep networks that age faces – generative AI from the era when GANs were the hottest thing around.

Even Klemsdal · Adam Gaidi

Cogito's very own reinforcement learning framework – the foundation many of the later RL projects were built on.

Jonathan Jørgensen · Marius Landsverk · Kim Midtlid · Even Klemsdal