Ai & Web Programming Challenge

Challenge 1:

A Heritage Brought Back to Life

Category: Deep Learning | Computer Vision | Neural Style Transfer

Problem & Project Description

A digital museum has a collection of historical images and valuable artworks. The goal of this challenge is to recreate images using different artistic styles while preserving the original content and structure. The system receives a Content Image and a Style Image. It must preserve the content of the original image while transferring the visual characteristics of the reference style. For example, a real photograph of a street should be transformed into an image that preserves the same objects and structure while adopting the visual appearance and texture of a painting.

Project Objective

Develop a system that addresses the core problem of “Challenge 1: A Heritage Brought Back to Life” and produces the defined output, with a focus on practical usefulness and reliable performance.

Expected Output

Expected Results / Competition Expectations

The solution should address the stated problem coherently, provide clear and useful outputs, and remain reasonably effective on new examples.

Challenge 2:

The Duplicate Question Detective

Category: Machine Learning | Natural Language Processing (NLP) | Semantic Similarity

Problem & Project Description

A large question-and-answer platform receives thousands of questions every day. Many of these questions have similar meanings but are written using different words and sentence structures.

For example:

“How can I learn Python programming?” “What is the best way to learn Python?” “Where should I start if I want to learn Python?” These questions are not identical at the word level, but they share a similar intent and meaning.

Teams must design a system that receives two questions and determines whether they are semantically similar or not. The system must distinguish between: Similar:

The two questions share a similar meaning or intent. Different:

The two questions have different meanings or intents. Simple word matching is not sufficient.

Two questions may contain very few common words while still expressing the same idea. Therefore, the model must learn or generate a meaningful semantic representation of each question and use it to determine their similarity.

Project Objective

Develop a system that addresses the core problem of “Challenge 2: The Duplicate Question Detective” and produces the defined output, with a focus on practical usefulness and reliable performance.

Expected Output

For each question pair, the system should provide at least: A Similar / Different prediction A similarity score or probability, if applicable The ability to process previously unseen questions without manual tuning for individual samples

Expected Results / Competition Expectations

The solution should address the stated problem coherently, provide clear and useful outputs, and remain reasonably effective on new examples.

Challenge 3:

AI Tutor — The Intelligent Teacher

Category: Artificial Intelligence | NLP | Information Retrieval | Semantic Search | Question Answering | RAG

Problem & Project Description

An educational platform allows students of different ages and educational levels to ask questions.

However, the same question should not necessarily receive the same type or level of explanation from the system.

For example, the question “What is a cell?” should be explained in a simple and understandable way to a 10-year-old student, while a 17-year-old student may require a more detailed and technical explanation.

The goal of this challenge is to build an adaptive AI Tutor that can understand the student’s question, retrieve relevant educational content, and generate an answer appropriate for the student’s level.

The system should perform the following process: Question → Understanding → Relevant Content → Level-Aware Answer → Source It must understand the question, retrieve the most relevant educational content, and generate an answer appropriate for the student’s level.

The system should not simply function as a general-purpose chatbot.

The answer should be grounded as much as possible in the provided educational resources, and the system must identify the source or relevant section used to generate the answer.

Project Objective

Develop a system that addresses the core problem of “Challenge 3: AI Tutor — The Intelligent Teacher” and produces the defined output, with a focus on practical usefulness and reliable performance.

Expected Output

Question ↓ Student Level ↓ Relevant Educational Content ↓ Adaptive Answer ↓ Source / Reference

Expected Results / Competition Expectations

The solution should address the stated problem coherently, provide clear and useful outputs, and remain reasonably effective on new examples.

Challenge 4:

Truth Guardian — AI Hallucination Detection & Reliability Assessment

Category: Generative AI | RAG | NLP | AI Evaluation | Hallucination Detection | Agentic AI

Problem & Project Description

Today, RAG and Agentic AI systems are widely used for answering questions and performing intelligent tasks.

However, a major challenge remains: an AI system may confidently generate information that is not supported by the retrieved sources, is partially missing from the available evidence, or directly contradicts the provided documents. In this challenge, teams take the role of an AI Evaluation & Safety Team and must assess the reliability of AI-generated answers.

Each sample may contain:

Question → Retrieved Documents → AI Answer → Evidence → Agent Logs The team must determine whether the generated answer is properly supported by the available evidence or contains a Hallucination.

In the advanced stage, teams must also predict the hallucination risk of each response. Hallucination detection is not always a simple text comparison task. More challenging examples may include: Answers containing both correct and incorrect information. Answers that use only part of the retrieved evidence. Claims that are not supported by any retrieved document. Retrieved documents with low relevance or insufficient evidence. Multi-step Agent workflows that generate or revise information over time. Therefore, the system must analyze the relationship between the question, retrieved documents, evidence, and final answer.

Project Objective

Develop a system that addresses the core problem of “Challenge 4: Truth Guardian — AI Hallucination Detection & Reliability Assessment” and produces the defined output, with a focus on practical usefulness and reliable performance.

Expected Output

The system should provide at least: Grounded / Hallucinated A hallucination risk score or probability Identification of suspicious or unsupported parts, when possible Evidence or an explanation supporting the system’s decision

Expected Results / Competition Expectations

The solution should address the stated problem coherently, provide clear and useful outputs, and remain reasonably effective on new examples.

Challenge 5:

Intelligent Visual Search Engine

Category: Computer Vision | Deep Learning | Transfer Learning | Image Embedding | Image Retrieval

Problem & Project Description

In this challenge, teams will receive a collection of images and must build an Image Retrieval system capable of finding the most visually similar images in the dataset for a new Query Image.

For example, if the query image contains a hamburger, the system should retrieve and rank the images in the dataset that are visually most similar to the query.

The system should not simply classify the image or predict its category.

Instead, it must answer: “Which images in the dataset are most visually similar to the Query Image?” The overall pipeline should follow: Query Image → Feature Extraction → Image Embedding → Similarity Search → Ranking → Top-K Results Similarity should not be determined solely through direct pixel-level comparison.

The system should extract meaningful visual features such as: Shape and structure Color Texture Visual patterns Higher-level image features These features should then be represented as an Embedding Vector. Images can then be compared and ranked based on the distance or similarity between their embeddings.

Project Objective

Develop a system that addresses the core problem of “Challenge 5: Intelligent Visual Search Engine” and produces the defined output, with a focus on practical usefulness and reliable performance.

Expected Output

Query Image ↓ Image Embedding ↓ Similarity Search ↓ Ranked Results ↓ Top-K Similar Images + Similarity Score

Expected Results / Competition Expectations

The solution should address the stated problem coherently, provide clear and useful outputs, and remain reasonably effective on new examples.

Challenge 6:

Intelligent Facial Landmark Mapper — Facial Keypoint Detection and Localization

Category: Computer Vision | Deep Learning | CNN | Facial Landmark Detection | Regression

Problem & Project Description

In many Computer Vision applications, detecting the presence of a face is not enough.

More detailed analysis requires identifying the precise locations of important facial features.

In this challenge, teams must build a model capable of detecting and localizing Facial Keypoints / Landmarks from face images, including: Eyes Eyebrows Nose Mouth corners Facial boundary points The model receives a face image and predicts the coordinates of its facial keypoints.

Some final test images may contain conditions that are less represented in the training data, such as: Different face angles Low-light conditions Image noise Different face scales Different face positions Different facial expressions Therefore, the model must be able to generalize beyond the patterns observed in the training data.

Project Objective

Develop a system that addresses the core problem of “Challenge 6: Intelligent Facial Landmark Mapper — Facial Keypoint Detection and Localization” and produces the defined output, with a focus on practical usefulness and reliable performance.

Expected Output

Expected Results / Competition Expectations

The solution should address the stated problem coherently, provide clear and useful outputs, and remain reasonably effective on new examples.

Challenge 7:

Age Intelligence — Intelligent Age Estimation and Facial Robustness

Category: Computer Vision | Deep Learning | CNN | Age Estimation | Regression | Classification

Challenge Description

Develop an intelligent system that estimates a person’s approximate age from a facial image and, optionally, assigns the person to an age group.

The system should go beyond producing a single age prediction. It should also communicate the reliability of its estimate and demonstrate robust performance under different image conditions.

For each input image, the system should attempt to:

The system should be evaluated on previously unseen faces and challenging image conditions.

Problem & Project Description

The final system should provide the following outputs:

During the final evaluation, the model should be tested on challenging images involving:

The system should be designed to generalize beyond its training images rather than relying on memorized examples or incidental correlations.

Age Consistency

If multiple images of the same person are available, the system should be evaluated for consistency across different lighting conditions, facial expressions, and camera angles.

To ensure a fair evaluation, images of the same individual should not be unintentionally distributed across training and test sets when testing generalization to unseen people.

Project Objective

Develop a system that addresses the core problem of Challenge 7: Age Intelligence — Intelligent Age Estimation and Facial Robustness.

The objective is to estimate approximate age from facial images while evaluating prediction accuracy, uncertainty, and robustness under challenging conditions.

Expected Output

The system should provide:

Expected Results / Competition Expectations

The solution should demonstrate a functional age-estimation pipeline and provide measurable evidence of its performance.

Evaluation should consider:

Participants should explain their dataset, preprocessing method, model architecture, evaluation strategy, and limitations.

Challenge Question: Can your model estimate the approximate age of an unseen person from a facial image while maintaining reliable performance under challenging conditions?

Challenge 8:

Hidden Connections in Email Data

Category : NLP | Machine Learning | Data Mining | Information Retrieval | Graph Analysis | Anomaly Detection

Imagine a large organization with an archive containing thousands of emails.

These emails contain information about people, topics, events, and communication patterns, but many of these relationships are not directly visible.

Your goal is to build an Intelligent Email Analysis System capable of extracting meaningful information, patterns, and hidden connections from a large email archive.

The system should answer questions such as: Which people communicate with each other most frequently?

Which topics repeatedly appear across emails?

Which emails are semantically related?

How are people connected through email communication?

Which relationships or patterns appear unusual?

Design an executable pipeline that extracts both textual and communication information from an email archive and uses it to discover important relationships, topics, and patterns.

Problem & Project Description

Each team must provide an executable pipeline: Email Archive → Data Extraction → Text Processing → Entity Extraction → Embedding & Similarity → Topic Discovery → Communication Graph → Anomaly Detection → Hidden Connections The system must be capable of processing new data and producing the required competition outputs.

Project Objective

Develop a system that addresses the core problem of “Challenge 8: Hidden Connections in Email Data” and produces the defined output, with a focus on practical usefulness and reliable performance.

Expected Output

Expected Results / Competition Expectations

The solution should address the stated problem coherently, provide clear and useful outputs, and remain reasonably effective on new examples.

Challenge 9:

Road Intelligence — Lane Detection and Vehicle Position Estimation

Category:  Computer Vision | Image Processing | Video Analysis | Deep Learning

Challenge Description

Develop an intelligent vision system that analyzes images or video from a forward-facing vehicle camera to identify lane boundaries and estimate the vehicle’s position relative to the lane center.

Real-world road scenes present several challenges, including changing lighting conditions, shadows, faded lane markings, curved roads, nearby vehicles, and camera movement. The system should be designed to handle these variations and produce reliable visual estimates.

From road images or video frames, the system should be able to:

Problem & Project Description

For each input frame, the system should estimate and report:

The system should visualize the detected lane boundaries and estimated lane center on the original image. The estimated lane region should also be highlighted. If the system estimates the drivable area separately, that region should be visualized distinctly.

For video input, the system should process consecutive frames and apply suitable temporal filtering or tracking techniques to reduce unnecessary fluctuations in its predictions.

The implementation may combine image-processing techniques, geometric methods, deep learning, and temporal analysis, depending on the selected approach.

Project Objective

Develop a vision-based system that addresses the core problem of Challenge 9: Road Intelligence — Lane Detection and Vehicle Position Estimation.

The objective is to go beyond detecting visible road lines by estimating lane geometry and the vehicle’s approximate lateral position relative to the lane center, while maintaining reliable performance under varying road and imaging conditions.

Expected Output

The system should:

Expected Results / Competition Expectations

The solution should demonstrate a functional pipeline from road-image input to lane estimation and visualization.

Evaluation should consider:

Participants should explain their lane-detection method, how they estimate vehicle position, and how they handle missing or uncertain lane boundaries.

This challenge focuses on perception and estimation. The system is not required to control the vehicle or operate as a complete autonomous-driving system.

Challenge 10:

Air Writing — Intelligent Character Recognition

Category : Computer Vision | Hand Tracking | Image Processing | Deep Learning | Human-Computer Interaction

Challenge Description

In this challenge, users write letters or digits in the air using their fingers, without touching a keyboard, mouse, or touchscreen.

The system must track the user’s hand and fingertip through a camera, convert the fingertip movement into a suitable representation, and use a Deep Learning model to recognize the written character.

The system should be able to:

Problem & Project Description

The final system should provide at least:

The system should transform the recorded trajectory into a representation suitable for model input. Possible approaches include rendering the trajectory as a two-dimensional image or processing the sequence of fingertip coordinates.

Participants may use a pretrained hand-tracking solution, image-processing techniques, and a suitable Deep Learning architecture. The recognition model should be trained or adapted to work with air-written characters rather than assuming that a model trained only on conventional handwritten character images will perform reliably.

Project Objective

Develop an interactive character-recognition system that addresses the core problem of Challenge 10: Air Writing — Intelligent Character Recognition.

The system should combine hand tracking, trajectory processing, and Deep Learning to recognize letters or digits written in the air, with minimal manual intervention and reliable performance on new writing samples.

Expected Output

The system should:

Expected Results / Competition Expectations

The solution should demonstrate a complete and functional pipeline from camera input to character prediction.

Evaluation should consider:

Participants should describe their trajectory representation, recognition model, training data, and evaluation method. Recognition performance should be evaluated on independent test samples rather than only on the training data.

Challenge 11:

Puzzle Intelligence — Vision-Based Logical Puzzle Solving

Category : Computer Vision | Image Processing | Constraint Satisfaction | Algorithm Design | Deep Learning (Optional)

Challenge Description

Build an intelligent system that receives an image of a logical puzzle, analyzes its visual structure, extracts the relevant information, converts the puzzle into a formal representation, and solves it automatically.

The system should be able to:

The system should minimize manual intervention throughout the process, from image input to solution generation.

Problem & Project Description

The system should provide an end-to-end pipeline consisting of the following stages:

Puzzle Image → Image Preprocessing → Board and Element Detection → Information Extraction → Constraint Modeling → Puzzle Solving → Solution Output

The implementation should address both visual interpretation and logical reasoning. The selected solving algorithm must be appropriate for the puzzle type and its rules.

Participants may use computer vision, optical character recognition (OCR), deep learning, constraint satisfaction techniques, backtracking, or other suitable algorithms, depending on the requirements of the selected puzzle.

Project Objective

Develop a system that addresses the core problem of Challenge 11: Puzzle Intelligence — Vision-Based Logical Puzzle Solving by transforming a visual puzzle into a formal problem representation and automatically generating a valid solution.

The system should demonstrate reliable information extraction, correct interpretation of puzzle constraints, and successful problem-solving with minimal manual intervention.

Expected Output

The system should:

The challenge is not simply to find the answer to a puzzle. The system must interpret the visual input, reconstruct the logical problem, and solve it through an automated pipeline.

Expected Results / Competition Expectations

The solution should demonstrate a complete and functional pipeline from image input to solution output.

Evaluation should consider:

Participants should clearly state which puzzle types their systems support and explain their visual recognition, problem representation, and solving methods.

Challenge 12:

Doodle Jump AI — Intelligent Agent with Reinforcement Learning

Category: Reinforcement Learning | Game AI | Decision-Making | Deep Learning (Optional)

Challenge Description

In this challenge, participants must develop an intelligent agent capable of playing Doodle Jump autonomously, without direct human control.

The agent must observe the game state, analyze the character’s position and the locations of platforms, select appropriate actions, and improve its strategy through interaction with the game environment.

The goal is not to memorize a fixed sequence of actions, but to develop an agent that learns to make effective decisions and adapts to new platform arrangements and game situations.

Problem & Project Description

The final system should provide an autonomous agent capable of:

Participants may use a reinforcement learning algorithm, with deep learning techniques if needed. The agent’s observations, available actions, and reward function should be defined clearly.

Project Objective

Develop an intelligent agent that addresses the core problem of autonomous gameplay in Doodle Jump using reinforcement learning. The agent should learn an effective decision-making strategy through interaction with the environment rather than relying on a predefined sequence of movements.

Expected Output

The agent should be able to:

Expected Results / Competition Expectations

The solution should demonstrate effective autonomous gameplay and provide measurable evidence of learning and performance.

Evaluation should consider relevant metrics, including:

Participants should explain their algorithm, state representation, action space, reward function, and evaluation method. High performance in a single run alone should not be considered sufficient evidence of a reliable solution.

Challenge 13:

Fraud Intelligence — Intelligent Financial Fraud Detection

Category : Machine Learning | Fraud Detection | Imbalanced Classification | Feature Engineering | Anomaly Detection

Challenge Description

In this challenge, participants must develop an intelligent system for detecting suspicious and fraudulent financial transactions.

The system must analyze transaction-level information and distinguish between legitimate and fraudulent transactions based on relevant transaction characteristics and behavioral patterns.

A major challenge is severe class imbalance, where fraudulent transactions represent only a small proportion of the available data. Therefore, high accuracy alone does not indicate an effective fraud detection system, as a model may achieve high accuracy while failing to identify most fraudulent transactions.

The goal is to detect fraudulent transactions effectively while controlling false alerts and maintaining reliable performance on previously unseen data.

Problem & Project Description

Develop a fraud detection system that processes financial transaction data and produces:

The system should follow a predictive pipeline such as:

Transaction Data → Preprocessing → Feature Engineering → Model Training → Fraud Prediction

Participants may use class-balancing techniques, cost-sensitive learning, classification algorithms, or anomaly detection methods as appropriate for the available data and problem formulation.

The model should learn meaningful fraud patterns rather than simply memorizing training examples.

Project Objective

Develop a system that addresses the core problem of Challenge 13: Fraud Intelligence — Intelligent Financial Fraud Detection. The system should identify potentially fraudulent transactions, reduce missed fraud cases, control false alerts, and generalize effectively to previously unseen transactions.

Expected Output

A functional Fraud Detection System capable of analyzing financial transactions and identifying suspicious or fraudulent activity.

The system should demonstrate the following capabilities:

Expected Results / Competition Expectations

The solution should demonstrate effective fraud detection, appropriate handling of class imbalance, and reliable performance on unseen data.

Evaluation should prioritize metrics suitable for imbalanced classification, such as Precision, Recall, F1-score, Precision-Recall AUC (PR-AUC), and the false positive rate. Accuracy may be reported as a supplementary metric but should not be the primary measure of success.

The solution should also explain the trade-off between detecting more fraudulent transactions and generating false alerts, with results evaluated at an appropriate decision threshold.

Challenge 14:

CareerMatch AI — Intelligent Applicant-to-Job Matching

Category : NLP | Recommendation Systems | Semantic Matching | Information Extraction | Machine Learning

Challenge Description

In this challenge, participants must develop an intelligent system that matches job applicants with suitable job opportunities.

The system should go beyond simple keyword matching and identify semantic relationships between a candidate’s skills, education, work experience, job titles, areas of specialization, and location, and the requirements of available positions.

The goal is to identify relevant job opportunities, estimate how well each candidate matches a position, and determine the gaps between the candidate’s current profile and the requirements of target roles.

Problem & Project Description

Develop an AI-powered career matching system that analyzes candidate profiles and job descriptions and provides:

Project Objective

Develop a system that addresses the core problem of Challenge 14: CareerMatch AI — Intelligent Applicant-to-Job Matching. The system should combine semantic understanding, information extraction, and recommendation techniques to identify suitable job opportunities, explain compatibility, and help applicants understand how to improve their qualifications for target roles.

Expected Output

A functional AI Career Matching System capable of analyzing candidate profiles and job-market requirements to deliver intelligent, explainable job recommendations beyond simple keyword-based search.

The system should demonstrate the following capabilities:

Expected Results / Competition Expectations

The solution should demonstrate relevant job recommendations, meaningful compatibility scores, and clear explanations of the factors influencing each match. Skill gap analysis should distinguish between mandatory and preferred qualifications when this information is available.

The system should be evaluated on previously unseen candidate profiles and job descriptions, using appropriate ranking and recommendation metrics where applicable. Its recommendations should be relevant, interpretable, and consistent with the information provided in the candidate’s profile and job requirements.

Challenge 15 :

Air Traffic Intelligence — Intelligent Taxi-Out Time Prediction

Category : Data Science | Machine Learning | Regression | Time-Series Analysis | Predictive Modeling

Challenge Description

In this challenge, participants must develop an intelligent system for predicting aircraft Taxi-Out Time — the time an aircraft spends moving on the airport surface from the start of taxiing until takeoff.

Taxi-Out Time depends on multiple factors, including airport traffic, flight schedules, airport layout, assigned runway, weather conditions, and operational procedures.

The goal is to learn patterns from historical flight data and predict Taxi-Out Time for new flights. The model should account for changing airport conditions and maintain reliable performance during both normal and congested periods.

Problem & Project Description

Develop a predictive system that processes flight data and relevant airport conditions to estimate the Taxi-Out Time for each flight.

The system should produce:

Project Objective

Develop a system that addresses the core problem of Challenge 15: Air Traffic Intelligence — Intelligent Taxi-Out Time Prediction. The system should learn from historical flight and airport data to estimate taxi-out duration for new flights and maintain reliable predictive performance under varying traffic and operational conditions.

Expected Output

A functional Airport Taxi-Out Prediction System with the following predictive pipeline:

Flight Data + Airport Conditions → Feature Engineering → Machine Learning → Taxi-Out Time Prediction

The system should demonstrate the following capabilities:

Expected Results / Competition Expectations

The solution should demonstrate accurate Taxi-Out Time predictions and reliable performance across different traffic conditions. Evaluation should use appropriate regression metrics, such as Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE).

Performance should be assessed separately for normal and congested conditions where sufficient evaluation data is available. The evaluation process must prevent data leakage by using only features available at the time the prediction is made.

Challenge 16:

Shipping Intelligence — Intelligent Delivery Time Prediction

Category : Data Science | Machine Learning | Regression | Logistics Analytics | Predictive Modeling

Challenge Description

In this challenge, participants must develop an intelligent system for predicting delivery times for logistics orders.

Delivery time depends on multiple factors, including origin, destination, route, transportation method, order time, shipment characteristics, and operational conditions. Traffic, weather, and unexpected delays may also affect delivery time when relevant data is available.

The goal is to learn patterns from historical shipping data and predict the delivery time of previously unseen orders.

Problem & Project Description

Develop a predictive system that processes historical order data and estimates the delivery time for each new order.

The system should produce:

Project Objective

Develop a system that addresses the core problem of Challenge 16: Shipping Intelligence — Intelligent Delivery Time Prediction. The system should learn from historical logistics data to predict delivery times for new orders and identify meaningful patterns that can support more efficient shipping operations.

Expected Output

A functional Shipping Delivery Time Prediction System with the following predictive pipeline:

Order Data → Data Analysis → Feature Engineering → Machine Learning → Delivery Time Prediction

The system should demonstrate the following capabilities:

Expected Results / Competition Expectations

The solution should demonstrate accurate and reliable delivery time predictions on unseen orders. Evaluation should use appropriate regression metrics, such as Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), or the coefficient of determination (R²).

The solution should also provide meaningful insights into factors associated with delivery delays, while avoiding unsupported conclusions about causation. Training and evaluation should be designed to prevent information from actual future delivery outcomes from leaking into the model’s input features.

Challenge 17:

Customer Segmentation Intelligence

Category : Machine Learning | Customer Analytics | Classification | Clustering | Feature Engineering

Challenge Description

An automotive company plans to enter new markets.

In its existing market, customers are divided into several behavioral segments based on demographic, economic, and behavioral characteristics. Each segment requires a different business strategy.

The company now has information about a large number of new customers, but their actual segments are unknown.

The goal is to develop an intelligent Machine Learning system that learns meaningful customer patterns from historical data and assigns new customers to the most appropriate segment.

Depending on the available data, participants may use existing segment labels to train a classification model or apply clustering techniques to discover customer groups when reliable labels are unavailable.

Problem & Project Description

Develop a customer segmentation system that analyzes customer data and produces the following outputs for each new customer:

Project Objective

Develop a system that addresses the core problem of Challenge 17: Customer Segmentation Intelligence — Intelligent Customer Classification. The system should identify meaningful customer patterns, assign new customers to appropriate segments, and provide useful insights to support market analysis and business decision-making.

Expected Output

A functional Customer Segmentation System capable of analyzing customer characteristics and identifying or predicting customer segments.

The system should demonstrate the following capabilities:

Expected Results / Competition Expectations

The solution should demonstrate meaningful customer segmentation, clear interpretation of segment characteristics, and reliable performance on previously unseen customer data.

For classification-based solutions, evaluation should use appropriate metrics such as accuracy, precision, recall, and F1-score. For clustering-based solutions, participants should use suitable clustering evaluation methods and demonstrate that the discovered groups are meaningfully different and useful for customer analysis.

The final system should provide interpretable results that can help the company understand customer differences and make informed business decisions.

Challenge 18:

Building Segmentation Intelligence

Category : Computer Vision | Deep Learning | Semantic Segmentation | Image Processing

Challenge Description

Accurate building maps are essential for urban mapping, disaster management, infrastructure planning, and the analysis of residential areas.

In this challenge, teams must develop an AI system that processes aerial or drone images and identifies buildings at the pixel level using semantic segmentation.

Each pixel must be classified into one of two categories:

The system should generate a Building Segmentation Mask that distinguishes building regions from the surrounding background as accurately as possible.

Problem & Project Description

For each input image, the system must generate a segmentation mask that:

Project Objective

Develop a system that addresses the core problem of Challenge 18: Building Segmentation Intelligence — AI-Powered Building Extraction. The system should extract building regions from aerial imagery at the pixel level while maintaining reliable segmentation quality across different locations and image conditions.

Expected Output

A functional Building Segmentation System that receives an aerial or drone image and generates a corresponding pixel-level segmentation mask.

The solution should focus on:

Expected Results / Competition Expectations

The solution should demonstrate accurate building extraction, reliable boundary delineation, and effective detection of small buildings. Performance should be evaluated using appropriate segmentation metrics, such as Intersection over Union (IoU), Dice Score, and Precision and Recall.

The model should also be evaluated on previously unseen images to assess its ability to generalize beyond the training data.

Challenge 19:

City Pulse AI

Category : Machine Learning | Time Series | Forecasting | Feature Engineering | Anomaly Detection

Challenge Description

A smart city continuously collects data from different urban areas. Historical data may contain temporal patterns and early indicators of changes in transportation activity, environmental conditions, public service usage, and other aspects of urban life.

Sudden changes in these patterns may indicate that an area is moving toward an unusual condition. Identifying such changes can help city authorities improve monitoring, resource allocation, and urban planning.

The goal is to develop an intelligent system that learns from historical urban data to forecast future conditions in different areas and identify potentially unusual patterns.

The system should account for temporal dependencies, short- and long-term trends, and relevant external factors when available. Although future events cannot always be predicted accurately, the system should identify deviations from expected patterns and provide useful forecasts for decision-making.

Problem & Project Description

Develop an urban forecasting system that analyzes historical data and produces the following outputs for each future time period:

Project Objective

Develop a system that addresses the core problem of Challenge 19: City Pulse AI — Smart Urban Forecasting and Anomaly Detection. The system should learn temporal patterns from historical urban data, forecast future conditions, and identify potentially unusual deviations to support practical urban monitoring and planning.

Expected Output

A functional Machine Learning system that learns from historical patterns and predicts the future values or conditions of a selected urban variable.

The system should demonstrate the following capabilities:

Expected Results / Competition Expectations

The solution should demonstrate meaningful forecasts, appropriate analysis of temporal patterns, and reliable performance on unseen time periods. Forecasting quality should be evaluated using suitable metrics, such as MAE or RMSE, while anomaly detection should be evaluated separately when implemented.

The solution should clearly distinguish predicted values from observed values and account for uncertainty where applicable. Any claim of early warning should be supported by evidence that the system can identify relevant changes before they occur.

Challenge 20:

Smart Event Intelligence — Attendance Prediction and Capacity Management

Category : Machine Learning | Classification | Behavioral Analysis | Feature Engineering

Challenge Description

An organization regularly holds educational programs and events with limited capacity. However, the number of registered participants does not always match the number of people who actually attend.

Overestimating attendance can lead to unnecessary resource allocation and capacity management problems, while underestimating it can result in unused capacity and inefficient resource planning.

The organization wants to develop an intelligent system that estimates the probability of attendance for each registered participant before an event takes place.

Using information about participants, event characteristics, and historical attendance behavior, the model must predict whether each registered participant will attend the event.

Potential features may include registration time, event type, event schedule, registration status, and previous attendance records, when available.

Problem & Project Description

Develop an attendance prediction system that processes registration and event data and produces:

Project Objective

Develop a system that addresses the core problem of Challenge 20: Smart Event Intelligence — Attendance Prediction and Capacity Management. The system should use historical data and relevant participant and event features to predict attendance and support more effective capacity and resource planning.

Expected Output

A functional attendance prediction and capacity management system capable of estimating actual event attendance before the event takes place.

The system should provide attendance probabilities and predicted attendance statuses for individual registrations. It should also aggregate individual predictions to estimate the total number of attendees and help the organization make informed decisions about event capacity and resource allocation.

Expected Results / Competition Expectations

The solution should demonstrate meaningful attendance predictions, reliable probability estimates, and useful event-level attendance forecasts. Its performance should be evaluated on previously unseen registrations or events using appropriate classification metrics, such as precision, recall, and F1-score, as well as probability-based metrics when applicable.

The system should also demonstrate how its predictions can support capacity planning while accounting for uncertainty in actual attendance.

Challenge 21:

Human Activity Intelligence — Intelligent Human Activity Recognition

Category : Machine Learning | Time Series | Classification | Pattern Recognition

Challenge Description

Develop an intelligent system that recognizes a person’s physical activity automatically from motion data without requiring them to manually record their activity.

Different activities produce different movement patterns. For example, walking, sitting, running, and climbing stairs generate distinct patterns of motion. However, these patterns may vary depending on the person’s speed, movement intensity, and activity duration.

The goal is to build a Machine Learning model that learns these patterns from sequences of motion data and identifies the activity being performed.

Given a time window of movement data, the system must predict the corresponding activity class based on the observed motion patterns rather than relying on a single measurement.

Problem & Project Description

Develop a Human Activity Recognition system that processes time-series motion data, such as accelerometer and gyroscope measurements, and produces:

Project Objective

Develop a system that addresses the core problem of Challenge 21: Human Activity Intelligence — Intelligent Human Activity Recognition. The system should learn meaningful temporal patterns from motion data and classify physical activities accurately and reliably across different movement conditions.

Expected Output

A functional Human Activity Recognition system capable of analyzing sequences of motion data over time and predicting the corresponding physical activity.

The system should demonstrate that its predictions are based on movement patterns across a time window rather than on an isolated measurement. It should output the predicted activity and, when supported by the model, confidence scores or class probabilities.

Expected Results / Competition Expectations

The solution should demonstrate effective recognition of the defined activity classes, meaningful use of temporal motion patterns, and reliable performance on previously unseen data. Evaluation should consider appropriate classification metrics, such as accuracy, precision, recall, and F1-score, particularly when the activity classes are imbalanced.

Challenge 22:

Lost & Found Intelligence — Intelligent Matching of Lost and Found Items

Category: Machine Learning | Classification | Record Matching | Feature Engineering | Ranking

Challenge Description

Develop an intelligent system that matches lost-item reports with potentially related found-item reports.

For each Lost Item report, the system may receive several Found Item reports. It must:

This is not simply a text similarity problem. The system should consider textual descriptions, item characteristics, time, and location together to identify potential matches.

Problem & Project Description

Develop a Lost & Found matching system that processes lost-item and found-item reports and produces:

Project Objective

Develop a system that addresses the core problem of Challenge 22: Lost & Found Intelligence — Intelligent Matching of Lost and Found Items. The system should combine relevant information from lost and found reports to identify likely matches, prioritize promising candidates, and provide useful results for practical lost-and-found operations.

Expected Output

A functional Lost & Found Intelligence System that identifies potential connections between lost and found item reports and presents the most likely matches to the organization.

The system should demonstrate the following processing pipeline:

Expected Results / Competition Expectations

The solution should demonstrate meaningful matching predictions, sensible candidate rankings, and effective use of textual, categorical, temporal, and geographical features. It should be evaluated on previously unseen report pairs to assess its ability to identify genuine matches while minimizing false matches and missed matches. Predicted probabilities should be reasonably calibrated when probability estimates are provided.

Challenge 23:

Fashion Vision AI — Intelligent Garment Condition Inspector

Category : Pygame | Game Development | Artificial Intelligence | Pathfinding | Decision-Making

Challenge Description

Build a 2D game using Pygame in which the player moves through a city map and completes multiple delivery missions.

The game must also include intelligent non-player characters (NPCs) that make decisions based on the current state of the environment.

Examples of intelligent behavior include:

AI Component

The intelligent behavior can be implemented using lightweight AI techniques, such as:

The goal is not to build a complex Machine Learning model. Instead, participants should demonstrate how AI-based decision-making can be integrated into the actual game mechanics.

Problem & Project Description

Develop a playable game containing:

Project Objective

Develop a playable game that addresses the core problem of Challenge 24: The Last Delivery — Smart Routing and Decision-Making Game. The project should demonstrate practical applications of intelligent routing, dynamic path selection, and decision-making in a changing game environment.

Expected Output

A functional 2D game in which AI is integrated into the game logic rather than implemented as a separate model.

The game should demonstrate the following components:

Expected Results / Competition Expectations

The solution should demonstrate functional pathfinding, meaningful AI-based decisions, and appropriate responses to changing environmental conditions. NPCs should behave consistently with their defined rules, and the game should remain playable across different mission configurations and obstacle arrangements.

Challenge 24:

The Last Delivery — Smart Routing and Decision-Making Game

Category : Pygame | Game Development | Artificial Intelligence | Pathfinding | Decision Making

Build a 2D game using Pygame in which the player moves through a city map and completes multiple delivery missions.

The game must also include intelligent non-player characters (NPCs) that can make decisions based on the current state of the environment.

For example: A pedestrian may choose a safer route.

A vehicle may avoid a blocked road.

A high-priority mission may be selected first.

An NPC may change its route when the environment changes.

AI Component The intelligent behavior can be implemented using lightweight AI techniques such as: Pathfinding A* Algorithm State Machines Rule-Based Decision Making Simple Scoring Systems The goal is not to build a complex Machine Learning model. Instead, participants should demonstrate how intelligent decision-making can become part of a real game.

Problem & Project Description

A playable game containing: A city map and game environment A controllable player Multiple delivery missions and destinations Obstacles and alternative routes NPCs with intelligent behavior A pathfinding system Score and time tracking Changing game conditions

Project Objective

Develop a system that addresses the core problem of “Challenge 24: The Last Delivery — Smart Routing and Decision-Making Game” and produces the defined output, with a focus on practical usefulness and reliable performance.

Expected Output

Expected Results / Competition Expectations

The solution should address the stated problem coherently, provide clear and useful outputs, and remain reasonably effective on new examples.

Challenge 25:

Robo Collector — The Smart Collector Robot

Category : Game Development | Pygame | AI Decision-Making

Challenge Description

Build a 2D game using Pygame in which the player and an AI-controlled robot compete to collect batteries and earn points.

The AI-controlled robot must decide which battery to target based on information such as:

The robot should continuously evaluate the game state and update its target when necessary.

AI Component

Machine Learning is not required. A rule-based decision-making system is sufficient.

The AI should follow a decision-making process such as:

Game State → Evaluate Batteries → Select Target → Move Toward Target

Participants may use distance calculations, simple scoring rules, or conditional logic to determine the robot’s next action. The robot should also avoid obstacles when moving toward its target.

Problem & Project Description

Develop a playable 2D game containing:

Project Objective

Develop a playable game in which an AI-controlled robot independently evaluates the current game state, selects appropriate targets, and makes decisions to collect batteries efficiently while avoiding obstacles.

Expected Output

A simple but complete Pygame game in which the AI-controlled robot can:

Expected Results / Competition Expectations

The solution should demonstrate a functional game with clear scoring and timing mechanisms, effective AI decision-making, and appropriate target selection. The robot should respond to changing game conditions and perform reasonably well across different game scenarios.