As part of my ongoing efforts to mentor future innovators in computer science and artificial intelligence, I’m sharing a curated set of project ideas. I invite students, educators, and collaborators to explore these ideas, adapt them, and build upon them. If you develop one of these projects or would like to collaborate, feel free to reach out through the contact form on this site.
1. Escape the Facility – Stealth Patrol Game
Concept:
The player must escape from a heavily guarded facility by avoiding detection. Guards patrol designated paths, investigate suspicious activity, and chase the player when spotted.
Topics Covered:
Search and Pathfinding (A* for guard patrols and chase behavior)
Perception Systems (FOV cones, sound-based detection)
Finite State Machine or Behavior Tree (Patrol → Investigate → Chase → Return)
Optional Extensions:
Procedural map generation
Adaptive patrols based on player success or patterns
Learning Focus:
Students build reactive, believable AI using layered perception and state-driven transitions.
2. Swarm Tactics – Horde Arena AI
Concept:
In an arena, AI-controlled enemies swarm and adapt to player behavior using group tactics such as flanking, retreating, or surrounding.
Topics Covered:
Pathfinding (multi-agent navigation, local avoidance)
Utility-based Decision Systems
Player Modeling (alter tactics based on player's behavior)
Optional Extensions:
Procedural enemy wave scaling
Machine learning for coordinated group movement
Learning Focus:
Students design multi-agent coordination with reactive behaviors and adaptive combat logic.
3. Dialoguemancer – NPC Social Puzzle Game
Concept:
NPCs in a fantasy town respond to the player through dynamic conversations that affect story progression and unlock social puzzles.
Topics Covered:
Dialogue Systems (rule-based or branching logic)
Social AI (trust, emotion, relationship modeling)
Behavior Trees (managing mood and social roles)
Optional Extensions:
Use of procedural dialogue generation
LLM-based text responses (optional)
Learning Focus:
Students learn to simulate social reasoning and dynamic conversational logic.
4. Wilderness Watch – Predator/Prey Simulation
Concept:
AI animals inhabit a forest and exhibit survival behaviors such as hunting, evading, and hiding, influenced by terrain, time, and player presence.
Topics Covered:
Perception Systems (sight, sound, smell)
Behavior Trees or FSM (hunt, flee, idle)
Procedural Terrain Awareness (movement and hiding zones)
Optional Extensions:
Adaptive AI that learns to avoid or pursue more efficiently
Machine learning-based evasion or detection behavior
Learning Focus:
Students simulate realistic animal intelligence with partial observability and environment interaction.
5. Adaptive Overlord – Boss Fight AI
Concept:
Design a boss that adjusts behavior across multiple fight phases, learning from the player's tactics and escalating difficulty over time.
Topics Covered:
FSM or Utility-Based AI (attack pattern control)
Player Modeling (detect tendencies and counter them)
Adaptive AI (phase transitions and behavior adjustment)
Optional Extensions:
Procedural attack pattern generation
Reinforcement learning for boss tuning
Learning Focus:
Students explore adaptive decision-making and dynamic difficulty scaling.
6. Procedural Dungeon AI Challenge
Concept:
A dungeon-crawling experience where layouts, puzzles, and enemy placements are procedurally generated. AI reacts intelligently to newly generated space.
Topics Covered:
Procedural Content Generation
Pathfinding (A* on dynamic maps)
FSM or Behavior Tree (enemy AI, traps)
Optional Extensions:
Dungeon-solving AI agent
Memory systems for map navigation
Learning Focus:
Students integrate dynamic environments with adaptable AI behavior.
7. Village Life Simulator
Concept:
NPCs live in a simulated village with individual schedules, emotions, and relationships. Player actions influence the social dynamics.
Topics Covered:
FSM or Behavior Tree (daily routines, social interactions)
Social AI (emotion graphs, reputation systems)
Procedural Environment Elements (weather, resource spawning)
Optional Extensions:
Player Modeling (NPCs remember past interactions)
Dynamic task assignment based on village needs
Learning Focus:
Students learn to coordinate multiple autonomous agents and simulate life-like behavior.
8. AI Escape Room Puzzle
Concept:
An AI agent (or the player) must navigate a puzzle-filled environment to escape, using logic, memory, and perception.
Topics Covered:
Search Algorithms (DFS, BFS, A*)
FSM or Behavior Tree for agent decision-making
Procedural Puzzle Generation
Optional Extensions:
Fog-of-war or partial observability
Memory systems to avoid previously failed paths
Learning Focus:
Students implement search logic, environmental interaction, and memory-driven exploration.
9. Zombie Survival Strategy Game
Concept:
Player must survive waves of intelligent zombies. AI adjusts its strategy based on noise, player movement, and past outcomes.
Topics Covered:
Pathfinding (dynamic movement through environments)
Perception Systems (sound-based detection, line of sight)
Adaptive AI or Player Modeling (escalating aggression)
Optional Extensions:
Procedural level and wave generation
Learning behavior based on prior encounters
Learning Focus:
Students build strategic AI that balances detection, adaptation, and challenge.
10. Interactive Theater – NPC Drama Engine
Concept:
NPCs participate in a scripted but reactive drama. Their actions change based on relationships, events, and player involvement.
Topics Covered:
Social AI and Dialogue Systems
Behavior Trees for scene and role control
Abstract Perception (trust, mood, alliances)
Optional Extensions:
Procedural narrative scenes
Visual social network graph of relationships
Learning Focus:
Students model AI that acts not just on stimulus, but on emotional and social context.
Description:
Transform Sokoban levels into tower defense maps, where the spatial layout of walls, paths, and crates defines enemy routes and placement zones.
Core Idea:
Convert crates into turrets, and pathways into enemy movement paths.
Player rearranges crates (pre-wave) using Sokoban-style movement to optimize defense.
Explore strategic implications of spatial reasoning from Sokoban applied to TD mechanics.
Potential Innovations:
AI-generated layouts from Sokoban levels with “difficulty curves” tied to path complexity.
Resource management layer linked to move efficiency.
Description:
Repurpose Sokoban levels into tactical RPG grid maps where solving movement-based challenges powers up your character or triggers events.
Core Idea:
Movement-based energy: push crates to charge spells or unlock areas.
Each puzzle becomes a “battle” or event in an RPG campaign.
Introduce enemies, traps, or environment changes based on Sokoban configurations.
Advanced Extensions:
Procedural RPG quest generation using Sokoban level constraints.
Difficulty scaling based on number of optimal pushes or number of movable objects.
Description:
Use Sokoban levels to generate factory layouts or resource pipeline constraints in a logistics/automation game.
Core Idea:
Treat crates as resources, goals as delivery spots, and levels as warehouse layouts.
Add optimization metrics: delivery time, energy used, space efficiency.
Use Sokoban levels as spatial challenges in a broader factory automation loop.
Possible Layers:
Integration with conveyor belts, loaders, and production chains.
Machine-learning agent for layout evaluation and optimization.
Description:
Turn Sokoban levels into detective game environments where pushing crates reveals clues, hidden paths, or suspects.
Core Idea:
Treat every Sokoban puzzle as a “scene” — solved correctly, it reveals narrative pieces.
Crates block access to crucial information; missteps obscure clues.
Players unlock character stories, items, or alibis through spatial logic.
Enrichments:
Random clue placement on crates/goals using Sokoban metadata.
Puzzle difficulty tied to clue importance.
Description:
Apply Sokoban principles to a space station or alien world, where pushing cargo/modules controls environmental states or energy flow.
Core Idea:
Convert crates into modular reactors or power units.
Solving movement puzzles reactivates ship systems, opens new regions, or prevents hazards.
Encourage strategic planning with consequences for inefficient moves.
Innovations:
Sokoban levels transformed into energy-routing puzzles (e.g., each goal lights up a section).
Real-time pressure (oxygen limits, threats) adds urgency to Sokoban problem-solving.
Description: Develop an intelligent system that allows users to upload a photo of a garment and receive recommended fabric patterns based on color harmony, cultural aesthetics, or fashion trends. The system then renders the selected pattern on the garment using a virtual try-on model (VITON-HD).
Deliverables:
Garment image uploader module
Fabric pattern recommendation engine
Pattern application using VITON-HD
User interface for preview and selection
Report on recommendation logic (e.g., color theory, aesthetic matching)
Final demo video and documentation
Resources:
VITON-HD GitHub: https://github.com/SUDO-AI/VITON-HD
OpenCV documentation: https://docs.opencv.org/master/
Fashion-MNIST dataset: https://github.com/zalandoresearch/fashion-mnist
Color theory in fashion: https://www.thedesignest.net/color-theory-guide/
Description: Create a fashion design generator where users describe garments in natural language, and the system uses an LLM (like GPT-4) to parse the style into attributes, then uses ControlNet and Stable Diffusion to render an image of the described design.
Deliverables:
Input text parser using OpenAI API
Structured attribute extractor (e.g., sleeve type, neckline, fit)
Image generation pipeline using ControlNet + Stable Diffusion
Optional: support for fabric pattern uploads
Final presentation, code repo, and report
Resources:
OpenAI GPT API: https://platform.openai.com/docs/introduction
ControlNet + Diffusers Colab: https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/controlnet.ipynb
Prompt Engineering Guide: https://github.com/dair-ai/Prompt-Engineering-Guide
Stable Diffusion Model: https://huggingface.co/CompVis/stable-diffusion-v1-4
Description: Build an AI chatbot that lets users describe garments or styling ideas (e.g., "a formal shirt with wax print sleeves") and renders them by combining text-to-image models with texture application tools.
Deliverables:
Chatbot interface (Streamlit or Gradio)
OpenAI integration for fashion term parsing
Texture blending module using OpenCV
Gallery of AI-generated designs
Evaluation of realism and user satisfaction
Resources:
Streamlit: https://streamlit.io/
Gradio: https://gradio.app/
OpenCV image blending tutorial: https://docs.opencv.org/4.x/d0/d86/tutorial_py_image_arithmetics.html
ChatGPT prompt examples: https://github.com/f/awesome-chatgpt-prompts
Description: Develop a system where users draw or upload sketches of garments and upload a fabric image to apply. The system detects garment regions and applies the pattern realistically using texture mapping and edge-aware blending.
Deliverables:
Sketch uploader and segmentation tool
Fabric pattern upload and preprocessor
Pattern mapping module (OpenCV/ControlNet)
Final try-on visualizations
Report detailing methods for fabric warping and masking
Resources:
Sketch2Image translation: https://github.com/hysts/pytorch_image_translation
OpenCV tutorials: https://docs.opencv.org/master/d9/df8/tutorial_root.html
DeepFashion2 dataset: https://github.com/switchablenorms/DeepFashion2
Description: Create a stylist assistant app where users upload wardrobe images and receive outfit suggestions powered by fashion rules and AI. Outfits can be styled with user-supplied or AI-generated fabrics.
Deliverables:
Responsive UI for image upload and selection
Outfit suggestion engine (LLM + image matching)
Fabric styling options (predefined or uploaded)
Optional mobile version or PWA
UX feedback collection and analysis
Resources:
OpenAI CLIP model: https://github.com/openai/CLIP
Streamlit for app prototyping: https://docs.streamlit.io/
Mobile PWA conversion: https://web.dev/progressive-web-apps/
Description: Develop a tool that scrapes public fashion content from social media platforms and builds a labeled dataset of garments with metadata like type, color, fabric, and occasion.
Deliverables:
Scraper for Pinterest or Instagram (respecting policies)
CLIP or LLM-powered auto-tagging module
Dataset format (CSV/JSON + image folder)
Dashboard for data filtering and export
Report on dataset quality and usage scenarios
Resources:
Scrapy framework: https://scrapy.org/
Pinterest scraping example: https://github.com/sdushantha/pinot
CLIP embedding tutorial: https://github.com/openai/CLIP
Hugging Face datasets guide: https://huggingface.co/docs/datasets/index
Description: Design a tool to promote cultural heritage through technology by allowing users to apply traditional textile patterns (e.g., Kente, Ankara) to modern garments, preserving design relevance and accessibility.
Deliverables:
Dataset of cultural textile patterns
Garment templates for pattern transfer
Interactive visualization tool
Research summary on cultural impact and ethical design
Live demo and presentation
Resources:
PatternBank archive: https://patternbank.com/
Meshroom for 3D modeling: https://alicevision.org/#meshroom
Cultural fashion design paper: https://www.tandfonline.com/toc/rffc20/current
Ethics in AI and fashion: https://dl.acm.org/doi/10.1145/3411764.3445744
Description: Develop a comprehensive tool that evaluates the quality of multiple-choice questions (MCQs) using AI. The system should assess clarity, distractor effectiveness, and cognitive complexity. Students will need to define a quality rubric, annotate a training set, extract linguistic and structural features, and experiment with machine learning models (e.g., BERT, SVM). The final product will include a web interface for batch evaluation, allowing instructors to upload questions and receive actionable insights.
Description: Create a system that organizes thousands of MCQs into logical groups and discovers latent topics using unsupervised learning. Students will employ techniques like sentence embeddings, K-means clustering, and BERTopic to extract meaningful clusters. A dashboard will allow instructors to explore clusters, filter questions by topic, and identify underrepresented areas. This project emphasizes NLP, data visualization, and interpretability.
Description: Design an adaptive quiz engine that suggests the next best question or topic based on a student's performance history. The system should incorporate student modeling techniques, collaborative filtering, or reinforcement learning. Students will implement backend logic, design a user interface for quizzes, and simulate or collect user data to evaluate recommendation accuracy. The project includes personalization logic, data handling, and interface design.
Description: Build a tool to analyze patterns in incorrect responses to identify common misconceptions in programming. The system will cluster similar wrong answers and tag them with conceptual misunderstandings (e.g., off-by-one errors, misuse of loops). Students will need to preprocess response data, use clustering and pattern recognition techniques, and validate their taxonomy with real users or instructors. The output should include visual analytics and targeted feedback generation.
Description: Develop an AI-powered assistant that reviews and suggests improvements to MCQs. The system will identify vague phrasing, poor distractors, or syntactic issues and propose revisions using LLMs. Students will implement both rule-based and AI-enhanced modules, create side-by-side comparison interfaces, and enable human-in-the-loop approval workflows. The goal is to streamline quiz curation and improve assessment quality.
Description: Create a predictive model that estimates the difficulty and cognitive level (e.g., Bloom's taxonomy) of MCQs. The project will involve defining difficulty metrics, extracting features (e.g., word complexity, distractor similarity), and training models on labeled data. Students will also build a web interface that lets instructors calibrate or filter questions by difficulty. This project combines machine learning, education theory, and applied NLP.
Description: Build a hybrid system that generates new MCQs from concept prompts and provides step-by-step explanations. Students will extract question templates from existing data, integrate GPT or similar models, and generate questions alongside rationales. The system should also support editing and review workflows for instructors. Emphasis is on explainability, template design, and AI integration.
Description: Design a tool that analyzes the question dataset for potential bias (e.g., cultural references, readability issues, gendered language). Students will implement natural language processing pipelines to detect bias, visualize metrics, and suggest inclusive alternatives. The final tool should include reporting features and allow instructors to review flagged content. This project combines ethics, NLP, and education.
Description: Build an intelligent tagging system that aligns MCQs to course outcomes, textbook sections, or specific concepts. Students will create or use an ontology of programming topics, train classifiers to tag questions, and build tools for manual correction. The system should support filtering, tag suggestions, and integration with quiz-building tools. This is a data-rich project with strong education-tech relevance.
Description: Develop a search engine that retrieves questions similar to a user’s query or concept input. Students will embed all questions using sentence transformers and use vector similarity search (e.g., FAISS). The interface should support keyword or example-based search and filters by topic, difficulty, or format. This project involves NLP, search engineering, and backend design.
Description: Create a system that accepts a code snippet and generates multiple-choice questions about its output, logic, or potential bugs. Students will combine static code analysis and AI-based generation techniques to identify key concepts and form relevant distractors. The system should be language-agnostic or support at least two programming languages. Includes code parsing, prompt engineering, and UI design.
Description: Design a tool that enables instructors to generate MCQs during live lectures based on spoken or typed input. Students will integrate speech-to-text or topic keyword entry with MCQ retrieval/generation tools. The final product should support real-time polling, student response collection, and visualization. This project merges live data processing, UX, and classroom technology.