Personal projects are a great way to learn new skills and explore your interests. Oftentimes, they make for very compelling extracurriculars, and can range from the creative arts to machine learning. There’s no better time to pursue them than when you’re in high school! If you are a high school student interested in starting an AI project, we have compiled a list of 10 AI project ideas for beginners. 

Why do an AI project in high school?

AI Projects are versatile, innovative ways to explore curiosities and address challenges. If you’re keen to learn about how AI can help address an issue, or if you’re generally passionate about learning, then the scientific nature and the possibilities of an AI project might be an experience you will benefit from. Not to mention – doing an AI project in high school can add an edge to your profile,  especially while applying to college, by demonstrating motivation and dedication to admissions officers. 

What are the best ways to start an AI project?

Starting an AI project can feel intimidating as a beginner, so here are some tips on the best way to start an AI project!

1. Start small because building your project at a small scale helps you learn and iterate quickly, as well as avoid a lot of complexities. Starting small helps smooth the process, and build a solid foundation before expanding your project!

2. Have a defined goal because a clear goal can become a very powerful guide to your project. While the ideas we’ve talked about in our blog are relatively clear, feel free to make them your own, and tailor them to your objectives.

3. Embracing the iterative nature of AI development can help finetune and improve your model and knowledge base. Examining all outcomes closely can help you determine whether you need to tweak hyper-parameters, or try different techniques!

Ultimately,  AI projects don’t just address real challenges – they also bring your imagined ideas to life. Going solo while learning AI can be challenging, so when you’re looking to start an AI project, don’t hesitate to reach out for mentorship or guidance! There are programs like the Veritas AI Fellowship that can help you learn and achieve your goals in your AI project. 

Here are some AI project ideas for beginners.

1. Training a Model for Handwritten Digit Recognition

Train a model to recognize handwritten digits! This project involves preprocessing the image data, implementing algorithms like CNN, and optimizing the model's accuracy. The first step for this project is to collect and preprocess the data – the good news for this project is that there are a lot of publicly available datasets for handwritten digits, like the MNIST dataset. This dataset actually contains 70,000 digits, all correctly labeled. Choose an algorithm to train your model in, then evaluate and optimize!

Level of Knowledge Needed: Beginner 

Skills Required: Machine learning, Python programming, familiarity with image processing libraries (e.g., OpenCV)

Coding Background Requirements: Basic understanding of machine learning and Python programming.

Potential Drawbacks: Obtaining a diverse and representative dataset of handwritten digits can be a challenge. The model may struggle with unusual or poorly written digits.

2. A Model for Image Classification

Build a model that can classify images into different categories (e.g., cats vs. dogs). It involves training a neural network on labeled image data and optimizing its performance. Some examples of this model include Google Cloud Vision API and Amazon Rekognition. 

Level of Knowledge Needed: Beginner

Skills Required: Machine learning, Python programming, familiarity with deep learning frameworks (e.g., TensorFlow, PyTorch)                                       

Coding Background Requirements: Proficiency in Python programming and understanding of machine learning concepts.              

Potential Drawbacks: Requires a good amount of labeled training data and computational resources for training complex models.

3. Spam Filtering

Try developing an AI system that filters out spam emails from a user's inbox. The outcome should be that the system analyzes the content of incoming emails and classifies them as spam or non-spam. Utilize techniques like text classification, feature engineering, and machine learning algorithms (e.g., Naive Bayes, Support Vector Machines) to build an effective spam filter. Note that once you’ve pre-processed your data, you want to extract identifiable features to classify data as spam. This could include specific words in the email, the length of the email, or the sender’s email address. At Veritas AI, we have had students work on similar projects with mentors from the University of Oxford! 

Level of Knowledge Needed: Beginner 

Skills Required: Natural Language Processing (NLP), Machine Learning, Python programming                              

Coding Background Requirements: Basic understanding of Python programming, NLP concepts, and machine learning algorithms.              

Potential Drawbacks: The effectiveness of the spam filter relies on the quality of the training data, the ability to handle evolving spam techniques, and avoiding false positives (legitimate emails classified as spam) or false negatives (spam emails bypassing the filter).

4. Developing Models for Sentiment Analysis

This project entails developing a model to analyze text sentiment (positive, negative, neutral), such as social media posts or customer reviews. It involves text preprocessing, feature extraction, and training a classification model. There are a lot of examples of this kind of model being deployed that are both commercial and open source, like spaCy and Amazon Comprehend. 

Level of Knowledge Needed: Beginner 

Skills Required: Natural language processing (NLP), Python programming

Coding Background Requirements: Basic understanding of Python programming and NLP concepts.

Potential Drawbacks: Accuracy can be affected by the quality and diversity of the training data, and the model may struggle with sarcasm or nuanced sentiment.

5. Models for Disease Prediction from Symptoms

Try building a model that predicts the likelihood of specific diseases based on input symptoms. Use labeled medical datasets and implement algorithms like decision trees or support vector machines. This model has been implemented in webapps like DiseasePrediction.ai and SymptomsChecker. In order to get the necessary dataset, it’s best to reach out to local hospitals and clinics – but you can also find public datasets on Medical Information Mart for Intensive Care (MIMIC-III) or ClinicalTrials.gov. With Veritas AI, students have previously worked on projects like these with mentors from Harvard and Cornell! 

Level of Knowledge Needed: Beginner to Intermediate

Skills Required: Machine Learning, Python programming

Coding Background Requirements: Basic understanding of Python programming and machine learning concepts.

Potential Drawbacks: Disease prediction accuracy depends on the quality and representativeness of the medical dataset used for training. False positives and false negatives can occur, leading to inaccurate predictions.

6. Making a Chatbot

Try creating a chatbot that can interact with users, provide information, or simulate conversations. This could involve processing user queries, understanding intents, and generating appropriate responses. Chatbots have been used in customer service, the healthcare industry, education, research, entertainment and more. In general, as the development of chatbots continues, they get used in many creative ways. So make sure you have a clear purpose for your chatbot!

Level of Knowledge Needed: Intermediate  

Skills Required: NLP, Python programming, conversational AI                                     

Coding Background Requirements: Proficiency in Python programming, and knowledge of NLP concepts and libraries.               

Potential Drawbacks: Developing a robust and natural-sounding chatbot can be challenging, and it may require more advanced NLP techniques for better understanding and context handling.

7. AI Systems for Object Detection

Build an AI system that can detect and classify objects in images or videos. It involves using pre-trained models, image preprocessing, and post-processing techniques.  An example of this system being executed can be found here, y a student from the Veritas AI fellowship! They made use of the CIFAR-10 dataset to recognize objects in different photos. 

Level of Knowledge Needed: Intermediate to Advanced

Skills Required: Computer vision, Python programming, familiarity with object detection models (e.g., YOLO, SSD)

Coding Background Requirements: Understanding of computer vision concepts, Python programming, and familiarity with deep learning frameworks.

Potential Drawbacks: Object detection can be computationally demanding and may require access to GPUs for real-time applications. Fine-tuning models for specific object classes can also be challenging.

8. Developing a Model for Music Generation

Try developing a model to generate music compositions. It involves preprocessing music data, training a generative model (e.g., RNN, GAN), and generating original music sequences. There are some open-source examples of this like Magenta. Some features you might want to look at when doing feature extraction from your dataset could include pitch, timbre, or duration. A veritas AI student previously worked on a similar project under the guidance of a PhD mentor from Columbia university. 

Level of Knowledge Needed: Intermediate to Advanced

Skills Required: Deep learning, Python programming, familiarity with music theory

Coding Background Requirements: Proficiency in Python programming, understanding of deep learning concepts, and familiarity with music-related libraries.

Potential Drawbacks: Generating high-quality and coherent music that matches human preferences and artistic style is a complex challenge. Music generation models may produce repetitive or unrealistic compositions.

9. Models for Language Translation

This project explores designing a model for translating text from one language to another. It involves preprocessing text data, training a sequence-to-sequence model, and handling language-specific challenges. There are a lot of examples of this type of model being deployed – like Google Translate and Microsoft Translate, but you can also find open source code as well. 

Level of Knowledge Needed: Intermediate to Advanced

Skills Required: NLP, Python programming, familiarity with sequence-to-sequence models                                      

Coding Background Requirements: Understanding of NLP concepts, sequence modeling, and Python programming.               

Potential Drawbacks: Translation accuracy can be affected by the availability of bilingual training data, handling rare words or idiomatic expressions, and preserving context in long sentences.

10. Building an extension for Deepfake detection

Try developing a Chrome extension that can detect deepfake videos or images while browsing! The extension should analyze media content and apply computer vision and machine learning techniques to identify potential deepfake manipulations. An example of such a model being executed can be found here, by a student from the Veritas AI fellowship!

Level of Knowledge Needed: Intermediate to Advanced  

Skills Required: Web development, computer vision, machine learning, Chrome extension development                                     
Coding Background Requirements: Proficiency in web development (HTML, CSS, JavaScript), computer vision algorithms, machine learning concepts, and Chrome extension development.               

Potential Drawbacks: Deepfake detection can be challenging due to the evolving nature of deepfake techniques, and achieving high accuracy can be difficult. Real-time detection and performance considerations can also pose challenges for the extension.

Note: It's crucial to have a strong understanding of deepfake detection techniques and be familiar with relevant libraries and frameworks. Keep in mind that deepfake detection is an active research area, and staying updated with the latest developments is essential to build an effective detection system.

If you’re looking to build unique projects in the field of AI/ML, consider applying to Veritas AI! 

Veritas AI was founded by Harvard graduate students, and through the programs, you get a chance to learn the fundamentals of AI and computer science while collaborating on real-world projects. You can also work 1-1 with mentors from universities like Harvard, Stanford, MIT, and more to create unique, personalized projects. In the past year, we had over 1000 students learn data science and AI with us. You can apply here!