Artificial Intelligence Engineering project ideas

100 projects, from a first build to a final-year capstone. Each lists the skills it teaches and the tools it needs; times and costs are estimates.

Beginner projects

• Linear Regression From Scratch — Fit a line with gradient descent and compare with the closed-form solution on a small dataset. Skills: regression, gradient descent. Tools: Python, NumPy. Data lab, Up to 15 h, Software only (free tools).
• Decision Tree Playground — Grow a decision tree on a toy dataset and watch splits change with depth. Skills: decision trees, overfitting. Tools: Python, scikit-learn. Visualizer, Up to 15 h, Under $30.
• K-Means Colour Quantizer — Reduce an image to k colours with k-means and compare quality across k. Skills: clustering, k-means. Tools: Python. Visualizer, Up to 15 h, Software only (free tools).
• Iris Flower Classifier — Train and evaluate a classifier on the Iris dataset with a confusion matrix. Skills: classification, evaluation. Tools: scikit-learn. Data lab, Up to 15 h, Under $30.
• Perceptron Visualizer — Show how a perceptron updates its boundary point by point. Skills: perceptrons, learning rules. Tools: web app. Visualizer, Up to 15 h, Software only (free tools).
• Spam Message Filter — Classify SMS messages as spam with Naive Bayes and inspect key words. Skills: text classification, Naive Bayes. Tools: Python. Data lab, Up to 15 h, Software only (free tools).
• Train-Test Split Explainer — Show why held-out data matters with an interactive overfitting demo. Skills: validation, generalisation. Tools: web app. Visualizer, Up to 15 h, Software only (free tools).
• Handwritten Digit Recognizer — Train a small network on MNIST and draw digits to test it. Skills: neural networks, MNIST. Tools: Python, PyTorch. Data lab, Up to 15 h, Software only (free tools).
• Movie Recommendation by Similarity — Recommend movies from user ratings with cosine similarity. Skills: recommenders, similarity. Tools: Python. Data lab, Up to 15 h, Software only (free tools).
• Sentiment Analysis of Reviews — Classify review sentiment and highlight influential words. Skills: sentiment analysis, NLP basics. Tools: Python. Data lab, Up to 15 h, Software only (free tools).
• Data Cleaning for ML — Clean a messy dataset, handle missing values and record each step. Skills: data cleaning, preprocessing. Tools: pandas. Workflow, Up to 15 h, Software only (free tools).
• Activation Function Explorer — Plot activations and their gradients and compare saturation. Skills: activation functions, gradients. Tools: web app. Visualizer, Up to 15 h, Software only (free tools).
• Rock-Paper-Scissors Pattern Bot — Predict an opponent's next move from history with a Markov model. Skills: Markov models, prediction. Tools: Python. Simulator, Up to 15 h, Software only (free tools).
• House Price Feature Study — Explore which features drive house prices with regression. Skills: feature analysis, regression. Tools: Python. Data lab, Up to 15 h, Software only (free tools).
• Image Augmentation Gallery — Apply flips, crops and colour jitter and see how datasets grow. Skills: data augmentation, visual explanation. Tools: Python. Visualizer, Up to 15 h, Software only (free tools).
• Chatbot With Rules and Intents — Build an intent-matching chatbot and log misunderstood inputs. Skills: intent matching, dialogue. Tools: Python. Workflow, Up to 15 h, Software only (free tools).
• Confusion Matrix Calculator — Compute precision, recall and F1 from a confusion matrix. Skills: metrics, evaluation. Tools: web calculator. Visualizer, Up to 15 h, Software only (free tools).
• Search Algorithm Maze Solver — Compare BFS, DFS and A* on mazes and count expanded nodes. Skills: search algorithms, heuristics. Tools: web app. Simulator, Up to 15 h, Software only (free tools).
• Tic-Tac-Toe Minimax Player — Implement minimax with alpha-beta pruning for tic-tac-toe. Skills: game trees, minimax. Tools: Python. Simulator, Up to 15 h, Software only (free tools).
• Word Embedding Explorer — Find nearest words and analogies with pretrained embeddings. Skills: embeddings, similarity. Tools: Python. Visualizer, Up to 15 h, Software only (free tools).
• Bias in a Toy Dataset — Show how skewed training data changes predictions for groups. Skills: fairness basics, bias. Tools: Python. Data lab, Up to 15 h, Software only (free tools).
• Teachable Image Sorter — Train an image classifier on your own photos with transfer learning. Skills: transfer learning, data analysis. Tools: Python. Data lab, Up to 15 h, Software only (free tools).
• Speech Command Recognizer — Recognise a few spoken commands from short audio clips. Skills: audio features, classification. Tools: Python. Data lab, Up to 15 h, Software only (free tools).
• Gradient Descent Learning Rate Demo — Show convergence and divergence for different learning rates. Skills: optimisation, learning rate. Tools: web app. Visualizer, Up to 15 h, Software only (free tools).
• AI Glossary Flashcards — Learn AI terms with spaced-repetition flashcards. Skills: concept learning, visual explanation. Tools: web app. Visualizer, Up to 15 h, Software only (free tools).

Intermediate projects

• Image Classifier With Explainability — Train a CNN and show Grad-CAM heatmaps explaining its predictions. Skills: CNNs, explainability. Tools: PyTorch. Data lab, 15–45 h, Software only (free tools).
• Tabular Model Comparison Lab — Compare logistic regression, random forest and gradient boosting with cross-validation. Skills: model selection, cross-validation. Tools: scikit-learn, XGBoost. Data lab, 15–45 h, $30–150.
• Time-Series Anomaly Detector — Detect anomalies in sensor series with rolling statistics and isolation forest. Skills: anomaly detection, time series. Tools: Python. Data lab, 15–45 h, Software only (free tools).
• Semantic Search Assistant — Search documents by meaning with sentence embeddings and a vector index. Skills: embeddings, vector search. Tools: Python, FAISS. Workflow, 15–45 h, Software only (free tools).
• Retrieval-Augmented Question Answering — Answer questions from your documents with retrieval and cited passages. Skills: RAG, retrieval. Tools: Python, LLM API. Workflow, 15–45 h, Software only (free tools).
• Small Language Model Evaluation Harness — Evaluate small language models on task sets with consistent prompts and metrics. Skills: LLM evaluation, benchmarking. Tools: Python. Data lab, 15–45 h, Software only (free tools).
• Reinforcement-Learning Control Sandbox — Train an agent to balance a cart-pole and plot learning curves. Skills: reinforcement learning, numerical modelling. Tools: Gymnasium. Simulator, 15–45 h, Software only (free tools).
• Multimodal Document Understanding — Extract fields from scanned forms combining OCR and layout. Skills: OCR, document AI. Tools: Python. Data lab, 15–45 h, Software only (free tools).
• Synthetic Data Quality Auditor — Compare synthetic and real data distributions and privacy risk. Skills: synthetic data, statistics. Tools: Python. Data lab, 15–45 h, Software only (free tools).
• Model Drift Monitoring Dashboard — Track input and prediction drift with PSI and alerts. Skills: drift detection, MLOps. Tools: dashboard. Dashboard, 15–45 h, $30–150.
• Hyperparameter Search Comparison — Compare grid, random and Bayesian search on the same budget. Skills: hyperparameter tuning, data analysis. Tools: Optuna. Data lab, 15–45 h, Software only (free tools).
• Object Detection on Custom Data — Train a detector on a labelled dataset and evaluate mAP. Skills: object detection, mAP. Tools: YOLO tooling. Data lab, 15–45 h, Software only (free tools).
• Text Summariser Comparison — Compare extractive and abstractive summaries with ROUGE and human ratings. Skills: summarisation, evaluation. Tools: Python. Data lab, 15–45 h, Software only (free tools).
• Customer Churn Predictor — Predict churn and explain drivers with SHAP values. Skills: classification, SHAP. Tools: Python. Data lab, 15–45 h, Software only (free tools).
• Image Segmentation for Defects — Segment surface defects with a U-Net and measure IoU. Skills: segmentation, U-Net. Tools: PyTorch. Data lab, 15–45 h, Software only (free tools).
• ML Experiment Tracker — Log runs, parameters and metrics to compare experiments reproducibly. Skills: experiment tracking, MLOps. Tools: MLflow. Workflow, 15–45 h, Software only (free tools).
• Named Entity Recogniser — Train an NER model for domain entities and evaluate per class. Skills: NER, sequence labelling. Tools: spaCy. Data lab, 15–45 h, Software only (free tools).
• Model Serving API — Serve a model behind a REST API with validation and latency logging. Skills: model serving, APIs. Tools: FastAPI. Workflow, 15–45 h, Software only (free tools).
• Recommendation With Matrix Factorisation — Factorise a rating matrix and evaluate top-k recommendations. Skills: matrix factorisation, data analysis. Tools: Python. Data lab, 15–45 h, Software only (free tools).
• Dimensionality Reduction Explorer — Compare PCA, t-SNE and UMAP projections on the same data. Skills: dimensionality reduction, visual explanation. Tools: Python. Visualizer, 15–45 h, Software only (free tools).
• Imbalanced Data Strategies Lab — Compare class weights, resampling and threshold tuning. Skills: imbalanced learning, data analysis. Tools: Python. Data lab, 15–45 h, Software only (free tools).
• Pose Estimation Fitness Counter — Count exercise repetitions from body keypoints. Skills: pose estimation, data analysis. Tools: MediaPipe. Data lab, 15–45 h, Software only (free tools).
• Prompt Engineering Workbench — Compare prompt variants for a task and record quality scores. Skills: prompt design, evaluation. Tools: LLM API. Workflow, 15–45 h, Software only (free tools).
• Speech-to-Text Accuracy Study — Measure word error rate across accents and noise levels. Skills: ASR evaluation, WER. Tools: Python. Data lab, 15–45 h, Software only (free tools).
• Genetic Algorithm Optimiser — Solve a scheduling problem with a genetic algorithm. Skills: evolutionary algorithms, numerical modelling. Tools: Python. Simulator, 15–45 h, Software only (free tools).

Advanced projects

• Federated Learning Simulator — Train a model across simulated clients with non-IID data and compare with central training. Skills: federated learning, non-IID data. Tools: Flower/PyTorch. Simulator, 45–100 h, Software only (free tools).
• Causal Inference Explorer — Estimate treatment effects with matching and DoWhy and test assumptions. Skills: causal inference, data analysis. Tools: Python, DoWhy. Data lab, 45–100 h, Software only (free tools).
• Graph Neural Network Demo — Classify nodes in a citation graph with a GNN. Skills: GNNs, graph learning. Tools: PyTorch Geometric. Data lab, 45–100 h, Software only (free tools).
• Active Learning Annotation Queue — Choose the most informative samples to label and track label savings. Skills: active learning, uncertainty sampling. Tools: Python. Workflow, 45–100 h, Software only (free tools).
• Edge Inference Latency Profiler — Profile model latency and memory on edge devices after quantisation. Skills: edge deployment, quantisation. Tools: ONNX Runtime. Data lab, 45–100 h, Software only (free tools).
• Privacy-Preserving Model Training Demo — Train with differential privacy and plot the accuracy-privacy trade-off. Skills: differential privacy, data analysis. Tools: Opacus. Data lab, 45–100 h, Software only (free tools).
• AI Agent Tool-Use Sandbox — Build an agent that calls sandboxed tools with logging and limits. Skills: LLM agents, tool use. Tools: Python, LLM API. Simulator, 45–100 h, Software only (free tools).
• Uncertainty Calibration Workbench — Measure calibration and apply temperature scaling. Skills: calibration, reliability diagrams. Tools: Python. Data lab, 45–100 h, Software only (free tools).
• Adversarial Robustness Benchmark — Evaluate robustness to perturbations defensively and test adversarial training. Skills: robustness evaluation, defence. Tools: PyTorch. Data lab, 45–100 h, Software only (free tools).
• LLM Fine-Tuning With LoRA — Fine-tune a small language model with LoRA and compare to the base. Skills: parameter-efficient fine-tuning, data analysis. Tools: PyTorch, PEFT. Data lab, 45–100 h, Software only (free tools).
• Vision Transformer From Scratch — Implement a small ViT and compare with a CNN. Skills: transformers, vision. Tools: PyTorch. Data lab, 45–100 h, Software only (free tools).
• Knowledge Distillation Study — Distil a large model into a small one and compare accuracy and speed. Skills: distillation, compression. Tools: PyTorch. Data lab, 45–100 h, Software only (free tools).
• Contrastive Self-Supervised Learning — Pretrain with SimCLR-style contrastive loss on unlabelled images. Skills: self-supervised learning, data analysis. Tools: PyTorch. Data lab, 45–100 h, Software only (free tools).
• Time-Series Transformer Forecaster — Forecast multivariate series with a transformer and baselines. Skills: sequence models, forecasting. Tools: PyTorch. Data lab, 45–100 h, Software only (free tools).
• Fairness Metrics Dashboard — Compare model performance across groups with fairness metrics. Skills: fairness, auditing. Tools: Fairlearn. Dashboard, 45–100 h, Software only (free tools).
• Diffusion Model Toy Implementation — Train a small diffusion model on simple images. Skills: generative models, diffusion. Tools: PyTorch. Data lab, 45–100 h, Software only (free tools).
• Multi-Armed Bandit Experimenter — Compare epsilon-greedy, UCB and Thompson sampling. Skills: bandits, exploration. Tools: Python. Simulator, 45–100 h, Software only (free tools).
• LLM Output Guardrail Tester — Test filters that block unsafe or off-topic outputs. Skills: guardrails, safety testing. Tools: Python. Workflow, 45–100 h, Software only (free tools).
• Model Compression Pruning Study — Prune networks and chart sparsity against accuracy. Skills: pruning, compression. Tools: PyTorch. Data lab, 45–100 h, Software only (free tools).
• Speech Synthesis Quality Evaluation — Compare TTS systems with MOS-style listening tests. Skills: TTS evaluation, data analysis. Tools: survey, Python. Data lab, 45–100 h, Software only (free tools).
• Document Chunking Strategy Study — Compare chunk sizes and overlaps for retrieval quality. Skills: retrieval tuning, data analysis. Tools: Python. Data lab, 45–100 h, Software only (free tools).
• Visual Question Answering Demo — Answer questions about images with a vision-language model. Skills: vision-language models, data analysis. Tools: Python. Data lab, 45–100 h, Software only (free tools).
• Anomaly Detection With Autoencoders — Detect anomalies from reconstruction error. Skills: autoencoders, data analysis. Tools: PyTorch. Data lab, 45–100 h, Software only (free tools).
• Planning With Monte Carlo Tree Search — Play a board game with MCTS and compare to minimax. Skills: MCTS, planning. Tools: Python. Simulator, 45–100 h, Software only (free tools).
• Data Labelling Quality Checker — Find label noise with cross-validation and confident learning. Skills: label quality, data analysis. Tools: cleanlab. Data lab, 45–100 h, Software only (free tools).

Expert projects

• Vision-Language Robotics Planner — Turn natural-language instructions into robot task plans in simulation. Skills: vision-language, task planning. Tools: simulator, LLM API. Simulator, 100 h+, $150+.
• Responsible AI Model Card Generator — Generate model cards with data, metrics, limits and intended use. Skills: model documentation, governance. Tools: Python. Workflow, 100 h+, Software only (free tools).
• LLM Prompt Regression Test Suite — Detect quality regressions when prompts or models change. Skills: LLM testing, regression. Tools: Python. Workflow, 100 h+, Software only (free tools).
• Forecasting Ensemble Workbench — Combine statistical and ML forecasts with stacking. Skills: ensembles, forecasting. Tools: Python. Data lab, 100 h+, Software only (free tools).
• Human Feedback Ranking Simulator — Simulate preference data and train a reward model. Skills: RLHF concepts, reward models. Tools: Python. Simulator, 100 h+, Software only (free tools).
• AI Deployment Canary Monitor — Route traffic to a new model gradually and roll back on regressions. Skills: canary releases, MLOps. Tools: dashboard. Dashboard, 100 h+, $150+.
• Neural Architecture Search Mini-Study — Search small architectures under a compute budget. Skills: NAS, efficiency. Tools: Python. Data lab, 100 h+, Software only (free tools).
• Multi-Agent Cooperation Simulator — Train agents to cooperate on a shared task. Skills: multi-agent RL, numerical modelling. Tools: PettingZoo. Simulator, 100 h+, Software only (free tools).
• World Model for Control — Learn a dynamics model and plan with it. Skills: model-based RL, numerical modelling. Tools: PyTorch. Simulator, 100 h+, Software only (free tools).
• Long-Context Retrieval Evaluation — Test how models use information across long inputs. Skills: long-context evaluation, data analysis. Tools: Python. Data lab, 100 h+, Software only (free tools).
• AI Energy and Carbon Tracker — Measure training energy and estimate emissions. Skills: green AI, measurement. Tools: CodeCarbon. Dashboard, 100 h+, Software only (free tools).
• Explainable Credit Decision System — Build an interpretable credit model with reason codes. Skills: interpretable ML, compliance. Tools: Python. Workflow, 100 h+, Software only (free tools).
• Continual Learning Benchmark — Measure forgetting when learning tasks in sequence. Skills: continual learning, data analysis. Tools: PyTorch. Data lab, 100 h+, Software only (free tools).
• Neuro-Symbolic Reasoning Demo — Combine a neural perceiver with logical rules. Skills: neuro-symbolic AI, numerical modelling. Tools: Python. Simulator, 100 h+, Software only (free tools).
• Neural Operator Surrogate Study — Train a Fourier neural operator as a fast surrogate for a family of PDE solutions and test generalisation. Skills: neural operators, scientific ML. Tools: PyTorch. Simulator, 100 h+, Software only (free tools).
• Medical Imaging AI Validation Study — Validate a model on external data with clear limits (research only). Skills: clinical validation, generalisation. Tools: Python. Data lab, 100 h+, Software only (free tools).
• Code Generation Model Evaluator — Evaluate generated code with unit tests and pass@k. Skills: code generation evaluation, data analysis. Tools: Python. Data lab, 100 h+, Software only (free tools).
• Knowledge Graph Question Answering — Answer questions by querying a knowledge graph from text. Skills: knowledge graphs, semantic parsing. Tools: Python, graph DB. Workflow, 100 h+, Software only (free tools).
• AI Red-Team Findings Tracker — Track defensive red-team findings and mitigations for a model. Skills: AI safety, risk tracking. Tools: database app. Workflow, 100 h+, Software only (free tools).
• Speculative Decoding Speed Study — Measure speed-ups from draft-model decoding. Skills: inference optimisation, data analysis. Tools: PyTorch. Data lab, 100 h+, Software only (free tools).
• Autonomous Driving Perception Stack Demo — Combine detection and tracking on a driving dataset. Skills: perception, tracking. Tools: Python. Data lab, 100 h+, Software only (free tools).
• AI-Driven Scientific Discovery Loop — Close the loop between model suggestions and simulated experiments. Skills: active learning, science. Tools: Python. Simulator, 100 h+, Software only (free tools).
• Mixture-of-Experts Toy Model — Implement routing among experts and study load balance. Skills: MoE, routing. Tools: PyTorch. Data lab, 100 h+, Software only (free tools).
• Model Supply Chain Provenance — Track datasets, weights and licences for released models. Skills: provenance, governance. Tools: database. Workflow, 100 h+, Software only (free tools).
• Real-Time Video Analytics Pipeline — Run detection on video streams with latency targets. Skills: streaming inference, KPI design. Tools: Python, GPU. Dashboard, 100 h+, Software only (free tools).

Project ideas for every branch · Artificial Intelligence Engineering roadmap