100 AI Capstone Project Ideas for IT Students (2026)

If your panel is approving more AI capstones than usual this year, you’re not imagining it. We’ve seen the trend across schools in Manila, Cebu, Davao, and a handful of universities in India that send us their syllabi. AI is the easiest category to get approved in 2026.

But it’s also the easiest to mess up.

The groups that fail their AI capstones almost always make the same mistake — the AI part of the project turns out to be a single ChatGPT API call wrapped in a Bootstrap form. The panel asks what model you trained, you say “we used the OpenAI API,” and the room gets very quiet.

This list is designed to avoid that.

Below are 100 AI capstone project ideas where the AI actually does something, grouped by technique so you can match them to the skills your team already has. We’ve marked the difficulty roughly, and many of the ideas link to source code we’ve already published.

Pick one. Build it. Don’t just demo it — explain what’s happening under the hood when you defend.

Why AI capstones are getting approved fastest in 2026

Panels are tired of “Online Ordering System with Database” titles. They’ve seen 50 of those this decade. When an AI proposal lands on their table, it sounds new — even when it isn’t, really.

That works in your favor. But only if you understand what panels actually want to see.

What panels want in an AI capstone this year

• A real model doing real work. Not a wrapper around someone else’s API. If you’re using a pretrained model, you better be fine-tuning it on your own data or chaining it with logic of your own.
• A measurable improvement over the non-AI version. If your project is “AI-powered attendance,” what’s better about your system compared to the spreadsheet the school is using now? Have numbers ready.
• A defensible dataset. Where did you get your data? How big is it? Did you clean it? Is it biased?
• Ethics addressed in writing. Privacy, consent, bias, misuse. One paragraph in Chapter 3 is enough, but you need that paragraph.

Three mistakes that kill AI capstones

• The “AI-powered” project with no actual AI. Calling something AI doesn’t make it AI. Calling openai.chat.completions.create() once doesn’t either.
• Using a pretrained model and not explaining what you did. “We used YOLOv8” is not a contribution. “We fine-tuned YOLOv8 on 800 photos of Philippine roads to detect potholes” is.
• No dataset documentation in Chapter 3. Your panel will ask. They always ask. Have the answer ready before they open their mouths.

Before you pick — what you actually need to finish an AI capstone

Don’t pick from the list yet. Read this first. It’ll save you a month.

Skills your team needs

• At least one person comfortable with Python and pandas. Not “took the intro course,” but “has built a notebook from scratch.” If your team needs a refresher, start with our Python tutorials.
• Someone who can put together a small web interface — Flask, Streamlit, or a simple React frontend. The AI is useless if your panel can’t see it work.
• Basic understanding of train/test split, accuracy, precision, and recall. If nobody on your team knows what a confusion matrix is, hold off on AI until they do.

Tools and hardware

You don’t need a gaming laptop. Most student AI projects fit comfortably inside Google Colab’s free tier, especially if you use the GPU runtime. For things Colab can’t handle, Kaggle Notebooks give you another 30 hours of GPU time per week, also free.

For pretrained models, use Hugging Face. For datasets, use Kaggle. For version control, use GitHub — and don’t skip this one. Panels in 2026 are starting to ask “where’s your repo?” If you can’t answer, you’re already behind.

The data problem nobody plans for

Most AI capstones don’t fail at the model. They fail at the data.

Three places to look:

• Kaggle has thousands of public datasets, many already cleaned. Start here.
• Public APIs — Twitter (now X), Reddit, news sites. Read the terms of service before scraping.
• Collect your own. This sounds like more work, and it is. But “we collected the dataset ourselves from 200 farms in Negros Occidental” is the kind of sentence that makes panels remember your project.

(See our free projects library for examples of how other student teams structured their datasets.)

You’ll probably need somewhere between 500 and 5,000 labeled examples to train a decent model on a domain-specific problem. If you’re fine-tuning a pretrained model, you can sometimes get away with fewer.

Computer Vision Capstone Ideas (20 ideas)

Computer vision projects are the most visual at defense — you can demo them live with the panel’s own faces, food, plants, whatever. That works in your favor, but it also means your demo has to actually work.

Browse our Python projects library for vision-based source code you can study first.

Beginner computer vision capstones

1. Face Recognition Attendance System — students walk in, camera logs them. Free source code available on this site.
2. Face Mask Detection — useful in clinics, schools, offices that still require masks.
3. Plant Identification from Photo — CNN classifier with mobile interface.
4. Garbage Classification System — biodegradable, recyclable, residual. Good for sustainability angle.
5. Fruit Quality Detection — ripe, unripe, spoiled. Easy to dataset yourself.
6. Handwritten Digit Recognition Web App — MNIST is overdone, but make the UI great and panels won’t care.
7. Animal Identification from Photo — useful for vets, wildlife groups.

Intermediate computer vision capstones

8. Plant Disease Detection for Farmers — train on PlantVillage dataset, deploy on phone. Strong agri-tech angle.
9. License Plate Recognition for Parking — OCR + detection. Useful for malls, gated villages.
10. Drowsy Driver Detection — webcam watches eyes, alerts when they close.
11. Pothole Detection from Dashcam — useful for LGUs, road inspection teams.
12. Helmet Detection for Motorcycle Riders — for traffic enforcement.
13. Crowd Counting System — count people in a frame. Useful for events, churches, transport.
14. Skin Disease Pre-Screening — needs heavy disclaimers. Don’t claim to diagnose.
15. Pose Estimation for Exercise Form — webcam checks if your squat is correct.

Advanced computer vision capstones

16. Object Detection for Traffic Monitoring with YOLO — count cars, motorcycles, pedestrians per minute.
17. Multi-Modal Sign Language Translator — Filipino Sign Language to text and speech.
18. Real-Time Falls Detection for Elderly — wall-mounted camera, instant SMS to family.
19. Forest Fire Detection from Satellite/Drone Images — Sentinel-2 data is free.
20. Image-to-Image Generation for Architecture — turn sketches into realistic renders. Heavy GPU needed.

Natural Language Processing Capstone Ideas (20 ideas)

NLP is the second easiest AI category to defend, especially in 2026 when LLMs blur the line between “you trained this” and “you fine-tuned a base model.” Don’t blur it on purpose. Be clear about what your contribution actually is.

Beginner NLP capstones

21. Sentiment Analysis Dashboard for Product Reviews — pull reviews from Shopee or Lazada, classify, visualize. Source code available.
22. Spam Email Classifier with Web Interface — train once, deploy with Flask.
23. Toxic Comment Detector for School Forums — pre-trained models work fine here.
24. News Article Auto-Summarizer — summarize Inquirer or Rappler articles into one paragraph.
25. Keyword Extractor for Research Papers — students paste a PDF, get a list of topics.
26. Email Auto-Categorizer — promotions, work, personal, spam.
27. Receipt OCR with Expense Categorizer — combines NLP with OCR.

Intermediate NLP capstones

28. Filipino Sentiment Analyzer — Tagalog and Bisaya tweets. The local-language angle is your differentiator.
29. Fake News Detector for Philippine Headlines — train on a mix of real and fact-checked false news.
30. Resume Screening AI for HR Departments — match resumes to job descriptions, rank candidates.
31. Tagalog-English Code-Switching Translator — Taglish in, clean English out (or vice versa).
32. Plagiarism Detector with Semantic Matching — beyond word matching, catches paraphrasing.
33. Customer Support Chatbot with Intent Classification — handles 80% of common questions.
34. Lecture Notes Auto-Summarizer from Recordings — speech to text, then summarize.
35. Job Description vs Resume Matcher — for both candidates and recruiters.

Advanced NLP capstones

36. Filipino Language Model Fine-Tuning Demo — fine-tune a small open model on Tagalog text.
37. Multi-Class Emotion Detector for Mental Health App — joy, sadness, anger, anxiety, etc.
38. Legal Document Clause Extractor — pull specific clauses from contracts.
39. Academic Paper Recommender from Abstract — students input topic, get 5 relevant papers.
40. Code Comment Auto-Generator — paste code, get docstrings.

LLM and Generative AI Capstone Ideas (15 ideas)

These are 2026’s hottest category. Panels love them. Just remember: if your “AI” is one API call to ChatGPT, you don’t have a capstone — you have a demo.

The trick to defensible LLM capstones is to add real engineering around the LLM. Retrieval, classification, agentic workflows, evaluation, guardrails. The model is the engine, not the car.

Educational LLM capstones

41. ChatGPT-Powered Tutoring Assistant for Specific Subjects — math, programming, biology. Subject-specific is more defensible than general.
42. RAG Document Q&A System for University Archives — students ask questions, the system answers using actual school documents.
43. AI Quiz Generator from PDF Notes — upload reviewer, get multiple-choice questions with answer keys.
44. AI Reading Companion for Grade Schoolers — reads aloud, asks comprehension questions.
45. AI Tutor with Conversation Memory — remembers what the student asked last week.

Productivity LLM capstones

46. AI Meeting Notes Summarizer — pulls action items, decisions, next steps.
47. AI-Powered Email Writer — draft replies in the user’s tone.
48. AI Lesson Plan Generator for Teachers — input topic + grade level, get a full lesson plan.
49. AI Resume and Cover Letter Builder — based on the job description.
50. AI Blog Post Outliner with SEO Suggestions — useful for small businesses.

Specialized agent capstones

51. LangChain-Powered Legal Document Analyzer — extract clauses, flag risky terms. Source code partial available.
52. AI Code Review Bot for Student Projects — checks for common bugs and style issues.
53. AI Customer Service Agent with Order Lookup — chats AND queries the database.
54. AI Recipe Generator from What’s in the Fridge — input ingredients, get dinner ideas.
55. Multi-Agent AI Research Assistant — one agent searches, another summarizes, a third writes the draft.

Recommendation System Capstone Ideas (10 ideas)

Recommendation systems are textbook ML projects. They’re well-documented, easy to dataset, and panels can immediately understand the use case.

56. Movie Recommendation System — collaborative filtering on MovieLens dataset.
57. Restaurant Recommender for Local Cities — based on cuisine, budget, distance.
58. Music Playlist Generator by Mood — mood detection + recommendation.
59. Book Recommender for School Libraries — based on past borrowings.
60. Tutor Matching System for University Students — match by subject, schedule, rating.
61. Job Recommender for Fresh Graduates — based on skills, location, salary preference.
62. Travel Destination Recommender — input budget and dates, get itinerary suggestions.
63. Online Course Recommender — based on career goals.
64. Personalized Diet Recommender — local Filipino food database makes this defensible.
65. E-Commerce Product Recommender for Local Shops — small Shopify stores will actually use this.

Predictive Analytics and Classic ML Capstones (15 ideas)

These are “traditional” ML projects — regression, classification, clustering. They feel less flashy than CV or LLM projects, but they’re the easiest to explain and defend in detail. If your team is more confident with pandas and scikit-learn than with neural networks, start here.

Business and operations

66. Customer Churn Prediction Dashboard — for telcos, ISPs, gyms, anyone with subscriptions.
67. Sales Forecasting for Small Retailers — last 12 months of POS data, predict next 3.
68. Employee Attrition Predictor — who’s likely to resign in 6 months.
69. Loan Default Risk Predictor — for community cooperatives and microfinance.
70. Dynamic Pricing Engine for Hotels or Resorts — based on demand, season, competitor rates.

Financial and market

71. Stock Price Prediction with LSTM — Source code available.
72. Cryptocurrency Trend Predictor — be careful with disclaimers.
73. Real Estate Price Predictor for Specific Cities — Manila, Cebu, Davao each have different models.
74. Insurance Claim Fraud Predictor — flag suspicious claims.
75. Bank Transaction Anomaly Detector — small, repeated transactions, weird hours.

Education and operations

76. Student Performance Predictor — at-risk students flagged early.
77. Crop Yield Predictor for Specific Regions — combine weather + soil data.
78. Flight Delay Predictor — useful for travelers and airlines.
79. Hospital Admission Volume Predictor — staffing tool for hospitals.
80. Energy Consumption Forecast for Schools — when to expect the highest electric bill.

Healthcare AI Capstones (10 ideas)

Healthcare AI is one of the most rewarding categories — and one of the riskiest. Panels (especially those with med-school connections) will scrutinize these hard. Always include disclaimers, never claim diagnostic capability, and ideally find a healthcare professional you can interview for Chapter 3.

81. Disease Symptom Pre-Screening Chatbot — informational only, refers to a doctor.
82. Pneumonia Detection from Chest X-Rays — train on the NIH Chest X-Ray dataset.
83. Diabetic Retinopathy Detection — retina images, severity grading.
84. Skin Lesion Pre-Screening App — flags suspicious moles for doctor follow-up.
85. Heart Disease Risk Predictor — based on standard clinical inputs.
86. Mental Health Companion App — chat-based, with safety guardrails and emergency hotline links.
87. Pill Identifier from Photo — useful for elderly with multiple prescriptions.
88. Pregnancy Health Tracker with AI Insights — milestones, warnings, doctor visit reminders.
89. Voice-Based Cognitive Health Tracker for Elderly — detects early signs of dementia.
90. Personalized Physical Therapy Routine Recommender — based on injury type and progress.

Speech and Audio AI Capstones (5 ideas)

91. Voice Assistant in Tagalog or Bisaya — Siri-style but for local languages. Wide open in 2026.
92. Lecture Note Generator from Audio Recordings — speech to text + summarization.
93. Music Genre Classifier — train on Spotify’s free audio features dataset.
94. Voice-Controlled Smart Home System — combines with IoT capstones nicely.
95. Audio-Based Cough Classifier — dry, wet, COVID-style cough detection. Heavy disclaimers required.

Security and Anomaly Detection AI Capstones (5 ideas)

96. Fraud Detection for Bank Transactions — train on Kaggle’s Credit Card Fraud dataset.
97. Network Intrusion Detection with ML — KDD or NSL-KDD dataset works well.
98. Phishing URL Detector with ML — beyond blacklists, uses pattern recognition.
99. Deepfake Image and Video Detector — celebrity face dataset for training.
100. Insider Threat Detector for Companies — anomalous employee behavior patterns.

How to defend an AI capstone

The defense for an AI project is different from a regular IS project. The panel will ask harder questions because AI sounds harder. Be ready.

Four questions almost every panel asks an AI group:

“Why did you choose this model?” Don’t say “because it had high accuracy.” Say “we compared logistic regression, random forest, and a small CNN. The CNN gave us the best F1 score on our validation set, so we went with it.” Have a comparison table in your appendix. Pair it with a data flow diagram showing how data moves from input to model to output — panels appreciate seeing the architecture, not just the metrics.

“Where did your data come from?” Be specific. “We used the PlantVillage dataset from Kaggle, plus 320 photos our team collected from farms in Bago City.” Bring the dataset summary slide.

“What about bias?” Every AI panel asks this now. Have a one-paragraph answer about who might be misrepresented by your dataset and what you’d do to fix it with more time.

“Couldn’t you just use ChatGPT for all this?” This is the killer question for LLM projects. Your answer has to show that your project does something a raw API call can’t do — retrieval, classification, integration with a database, multi-step reasoning, fine-tuned domain knowledge. Have that answer rehearsed.

If you can answer those four questions calmly, you’ll pass.

Pick one. Build it. Defend it.

The AI category is wide open in 2026 because most students are still doing PHP information systems. If your group has even one strong Python person, you have an advantage. Use it.

Browse our free Python projects for code you can study before you start. If your AI capstone needs UML diagrams (and it will — at least a use case and a sequence diagram), our UML guides cover every diagram type panels will ask about.

And if you haven’t picked a topic yet, the full list of 150 capstone ideas across all categories is here: 150 Best Capstone Project Ideas for IT Students 2026.

Now stop scrolling. Pick one. Open a notebook. Write the problem statement tonight.