VivaKitFinal-year AI projects you actually understand.
Computer vision, NLP, cybersecurity, IoT, energy, medical AI and more - each kit is a complete, readable Python project with a trained model, a demo app for your viva, honest results and a learning guide that explains every design choice. You learn it, extend it and write your own report.
35 projects across AI, vision, NLP, security and more
Pick a project to see what you will learn, our honest results and what is included. Screenshots are the real demo apps running our trained models.
Computer Vision 9 kits
Online Exam Proctoring & Cheating Detection Intelligence
Automated multi-modal computer vision and deep learning examination proctoring for academic institutions, certification bodies, and remote testing portals.
AI Assistive Vision & Real-Time Directional Voice Guidance
Real-time edge computer vision and directional audio mobility assistant for the visually impaired.
AI Yoga & Exercise Posture Correction Assistant
Real-time computer vision and kinematic biomechanics coaching engine powered by Google MediaPipe Pose Landmarker (Apache 2.0).
Driver Drowsiness Detection: EAR, PERCLOS, Yawns, Nods
Your project raises a red-screen-and-beep alarm when a webcam sees eyes closing, a yawn or a head nod.
Indian Traffic Sign Recognition System
Embedded-friendly computer vision system recognizing Indian Road Congress (IRC: 67-2012) regulatory, cautionary, and informatory traffic signs with Motor Vehicles Act statutory interpretations and Grad-CAM visual explanations.
Vehicle & Pedestrian Multi-Object Tracking & Counting
Real-time multi-class traffic surveillance and footfall analytics for smart cities and transportation engineering.
Waste Sorting with Deep Learning
Photograph one waste item (upload or webcam) and see whether it is cardboard, glass, metal, paper, plastic or trash, the top-3 guesses, the dry-waste bin it belongs in, and a heat-map of the evidence.
AI Virtual Mouse: Hand-Gesture Cursor Control
Your project moves the real cursor with your index finger, clicks when you pinch and scrolls with two fingers, with a fail-safe so it can never trap the mouse.
Devanagari Handwritten Character Recognition
Draw a letter with the mouse and a small CNN (PyTorch) reads it live: 46 classes, क to ज्ञ and the digits ० to ९, trained on 92,000 handwritten images.
Medical AI 5 kits
Bone Fracture Detection on X-rays
Faster R-CNN (torchvision, BSD licence - no AGPL code) draws boxes around suspected fractures on hand, leg, hip and shoulder X-rays.
Brain Tumour MRI: Segmentation & Classification
A multi-task U-Net that outlines the tumour and names its type (meningioma, glioma, pituitary) in one pass.
ECG Arrhythmia: Beat-by-Beat Classification
A 1D CNN with RR-interval features labels every heartbeat as normal, supraventricular, ventricular or fusion (AAMI classes) on the MIT-BIH Arrhythmia Database, with the inter-patient split, so no patient is in both training and test.
Retinal OCT: CNV, DME, Drusen or Normal
Four-class transfer learning on 83,484 retinal OCT scans, patient-wise splits, class weighting, macro-F1 and a screening view.
Pneumonia Detection from Chest X-rays
Transfer learning on 5,856 paediatric chest X-rays, patient-wise validation, Grad-CAM written from scratch, and a demo that shows why the model decided.
NLP 5 kits
Document OCR & Key Field Extraction
Digitize and extract structured business intelligence from Indian GST invoices, medical bills, and university tuition receipts.
Resume Screening & JD Matching
Paste a job description and get a ranked shortlist, with the required skills each resume has and lacks.
Clickbait & Misleading Headline Detection System
Production machine learning headline integrity analyzer and viral clickbait classifier.
SMS spam detection
Imbalanced classes done right: precision, recall and why 98% accuracy can mislead.
Sentiment analysis
TF-IDF text classification with explanations, and a cross-domain test (Amazon, IMDb, Yelp).
Classic ML 3 kits
Credit Card Fraud Detection
Spot the 75 frauds hidden among 56,962 real card transactions that come later in time than anything the models learnt from.
Heart disease prediction
Tabular classification on the UCI Cleveland data: pipelines, cross-validation, sensitivity vs specificity.
House price prediction
Regression with baselines, MAE vs RMSE, permutation importance and predicted-vs-actual plots.
Cybersecurity 3 kits
Android Malware Detection from Permissions
Static malware detection from the permissions an app requests, on 29,332 apps from the NATICUSdroid dataset (permission flags only - no malware is shipped).
IoT Network Intrusion Detection
Normal traffic or one of 9 attacks (SYN flood, Slowloris, ARP poisoning, SSH brute force, Nmap scans), predicted from the Zeek statistics of each network flow in the RT-IoT2022 dataset.
Phishing website detection
Full vs URL-only models: how much page and domain signals are worth, and spotting data-quality issues.
Environment & Energy 3 kits
Air Quality (PM2.5 / AQI) Forecasting
Forecast PM2.5 1 to 24 hours ahead, and the next-24-hour mean that India's National Air Quality Index uses, at 12 monitoring stations (4 years of hourly data).
Long-Range Monsoon Rainfall Prediction & Drought Risk Estimator
Long-range meteorological rainfall forecasting and agricultural drought classification engine powered by official 115-year India Meteorological Department (IMD) observational data (1901–2015).
Solar PV Power Output & Irradiance Forecasting System
Utility-scale day-ahead solar power generation and irradiance forecasting engine trained and evaluated on 2 full years (17,520 hours) of real hourly satellite data from NASA POWER for Bhadla Solar Park (Rajasthan, India), the world's largest solar installation.
Electrical (EEE) 2 kits
EV Li-Ion Battery State-of-Health & Remaining Useful Life Prognostics
Machine learning Battery Management System (BMS) prognostics engine powered by the official NASA Ames Prognostics Center of Excellence (PCoE) 18650 Li-ion battery run-to-failure dataset.
Transmission-Line Fault Detection, Classification & Localization
Physics-based protective relaying and transmission line diagnostic engine for 400 kV extra-high-voltage power networks.
Everything you need to learn, run, demo and extend
Readable source code
Python (PyTorch, scikit-learn and friends), split into small commented modules: data, model, training, evaluation, inference. One config file for every experiment.
Demo app for your viva
A polished local Gradio app with samples, explanations and a model card. Runs offline on your laptop - your data never leaves your computer.
Learning guide
Background on the problem, dataset pitfalls, every design choice explained, how to read the metrics, limitations, extension ideas and viva questions with answers.
Trained model + reference results
Weights we trained, plus confusion matrices, curves and example predictions - so you can compare your own runs fairly.
Dataset download script
Fetches the public dataset from its original source with checksum verification - no Kaggle account, no login - and prints the citation for your report.
One-click setup & tests
setup_windows.bat installs pinned packages (GPU build if you have NVIDIA, CPU otherwise).
Smoke tests prove the whole pipeline works on your PC in under a minute.
From purchase to your own experiment
Download & set up
Get the zip, double-click setup_windows.bat,
then run_demo.bat. The demo works before you download any data.
Learn & reproduce
Read the guide with the code open. Download the dataset, train and evaluate - and check your numbers against ours.
Extend & write it up
Pick an extension from the guide, run a fair comparison and write your own report about what you did and learned.
We sell code and teaching - not finished projects
Our promise and your responsibility
These kits are learning material, like a very detailed textbook with working code. Used well, they help you understand a real AI project deeply enough to extend it and defend it.
- We do not write reports, theses, synopses or presentations - not for money, not as a bonus. The kits contain no ready-to-submit documents.
- You must understand, extend and write up the work yourself. Each guide suggests extensions and ends with the question every examiner asks: "What did you add?"
- Acknowledge the kit and cite the dataset in your report, and follow your college's rules on external code. Ask your guide if you are unsure.
- Educational models. Not for medical, safety-critical or security decisions in real life.
One-time price, no subscription
One purchase covers one project team of up to four students at the same college.
Single kit
Pick the project that fits you
- Code, demo app, trained weights
- Learning guide with viva Q&A
- Setup script, tests, dataset downloader
- E-mail help with setup problems
Any three kits
Pick any three kits - mix tasks and datasets
- Everything in each of your three kits
- Compare tasks, metrics and architectures
- Great for teams still choosing a topic
- E-mail help with setup problems
Prices in Indian rupees. Digital download. See the refund policy.
Questions students ask
Can you write my project report, synopsis or PPT?
No. We only provide code and teaching material. Your report must be your own work - see the academic integrity section above.
Is the dataset included?
No - each kit includes a script that downloads the data from its original public source (no login) and a few samples so the demo works immediately. Webcam and microphone kits use data you record yourself. Every dataset's licence is checked, and its page tells you what to cite.
Do I need an NVIDIA graphics card?
Not for the demo, evaluation or tests - they run on any modern laptop CPU, and most kits also train on a CPU in minutes. Image-heavy kits train much faster on an NVIDIA GPU; each kit page says what it needs.
Will my results match yours exactly?
Close, not identical - GPUs and library versions introduce small random differences. The kit ships our metrics and the exact data split so you can compare fairly.
Can my whole team use one kit?
Yes: one purchase covers one project team of up to four students at the same college. Please don't share it beyond your team.
Can I change the code and add my own ideas?
Yes - that is the point. Every guide lists extension ideas from easy to hard, with where to start in the code.
Why are some of your numbers not "99 % accuracy"?
Because we split by patient, person, device or time and evaluate once on a held-out test set. Many online projects report inflated numbers caused by data leakage. Honest numbers, explained well, impress examiners more.
Which kit should I choose?
Start from your branch and interest: use the category filters above. Starter kits (₹1,499) are classic machine learning and gentle to learn; Standard and Advanced kits add deep learning, vision, signals or multi-step pipelines. Ask your guide which fits your syllabus.
Does it work on Linux, macOS or Google Colab?
The Python code is cross-platform; only the one-click .bat files are Windows-specific.
On other systems create a virtual environment and pip install -r requirements.txt.
We test and support Windows 10/11.
Can I use a kit in a real product, hospital or company?
No. These are educational models trained on small public datasets - not medical devices, security products or production systems.
How do I get the kit after paying?
Pay securely through Dodo Payments (UPI or card). You get an order confirmation by e-mail straight away, and your personalised download link within 24 hours (usually much sooner). For the three-kit bundle, reply to the confirmation with the kits you want. Problems? Write to [email protected].
Dataset attributions
Every kit uses public data whose licence allows this use, or data you record yourself. Each kit's page lists its dataset, the citation to put in your report and the licence. Screenshots show samples processed by our models; dataset authors do not endorse these kits.
Pick a project you can explain with confidence
Readable code, honest results and a guide that teaches - starting at ₹2,999.