FOOD INGREDIENT DETECTION & RECIPE RECOMMENDER | PYTHON, YOLOv8, T5, GRADIO:
Embedded Files
This project presents a two-stage AI system that transforms food image analysis by detecting multiple individual ingredients in a dish and recommending relevant recipes based on them. Using YOLOv8, the model achieved a mean Average Precision (mAP@50) of 0.829 for accurate, real-time multi-ingredient detection. The second stage leverages a TF-IDF-based retrieval system and a T5 model to suggest context-aware recipes tailored to the identified ingredients. Hyperparameter tuning was performed using Bayesian Optimization via Optuna. The solution is deployed with an interactive Gradio interface for real-time predictions and visualization
Yolov8 Demo.mp4DIABETES PREDICTION:
This project addresses the challenge of classifying diabetes stages using an imbalanced health dataset from Kaggle, where the target variable includes three classes: no diabetes (0), prediabetes (1), and diabetes (2). To tackle class imbalance, resampling techniques like SMOTE and ADASYN were applied along with dimensionality reduction via PCA. Multiple machine learning models—Random Forest, XGBoost, and neural networks—were used and optimized through Bayesian hyperparameter tuning. Resampling methods notably improved performance for underrepresented classes, and ensemble models benefited significantly from optimization. The project was implemented using Python with libraries like scikit-learn, imbalanced-learn, XGBoost, TensorFlow, and Optuna.
NO THIRD PARTY
NO THIRD PARTY
Fetch_stock_symbol.mp4AUTOPROPHET:
Fetch Stock Ticker Symbol w/o Third Party Applications (Left):
This aimed at identifying the stock names from user queries (in the form of sentences) and return their ticker symbol.
To achieve this, a dataset was downloaded from Kaggle. The stop words and other words which could create a bad match were removed from the user query. The best match was identified and the corresponding ticker symbol was returned
Fetch Stock Ticker Symbol using EDEN API (Right):
I applied prompt engineering to achieve the results of fetching stock ticker symbols from the user query using EDEN API.
The user query with multiple stock names was fed. The API correctly identified the stock name and correspondingly returned the ticker symbol for the particular stock.
EDEN API
EDEN API
Fetch_Symbol_EdenAPI.mp4DETECTING FRAUDLENT TRANSACTIONS:
Built a fraud detection system using real-world Kaggle data, applying PCA-transformed features and advanced resampling (ADASYN, undersampling) to train optimized KNN, SVM, Random Forest and XGBoost models.
Achieved up to 99% AUC and 87% F1-score on test data
Compared Decision Tree, Random Forest, XGBoost, KNN, and SVM
Implemented stratified cross-validation and GridSearchCV tuning
Addressed class imbalance using ADASYN and RandomUnderSampler
Page updated
Report abuse