Professional Projects
voice_GPT [GenAI, Conversational AI] [GitHub]
Built a production-ready, voice-controlled chatbot using OpenAI GPT and browser speech recognition, enabling seamless voice-to-text and contextual multi-turn conversations in enterprise and educational settings. Demonstrated sub-second response latency and flexible user interface for both voice and text input.
Impact: Rapid prototype for enterprise pilots; improved accessibility for AI-driven chat applications.
Problem: Achieving real-time, multi-turn voice and text interaction.
Solution: Integrated browser speech APIs and Python NLP pipelines for low-latency, enterprise-grade conversational UX.
Stack: Python, OpenAI GPT, transformers, SpeechRecognition, pyttsx3, Jupyter Notebook, JavaScript
Banana Plant Disease Classification [Agriculture, Computer Vision] [GitHub]
Implemented a deep transfer learning pipeline for early disease detection in bananas, supporting farmers and agri-tech partners. Achieved over 92% detection accuracy on field-validation sets.
Impact: Pilots suggest potential 20%+ reduction in crop loss for partner farms.
Problem: Limited, variable image data impacting accurate and scalable deployment.
Solution: Robust augmentation and fine-tuned CNNs; automated web interface for non-technical usage.
Stack: Python, Keras, TensorFlow, OpenCV, Transfer Learning
Healthcare AI (Brain Tumor Detection) [Healthcare, Medical Imaging] [GitHub]
Developed a CNN-based pipeline for MRI brain tumor diagnostics, overcoming data imbalance and boosting clinical AI sensitivity/specificity by 15%. Model reached 95.7% diagnostic accuracy on 50,000+ patient records.
Impact: Reduced diagnostic turnaround by 35%; improved clinical decision-making, adopted by radiology teams in pilot studies.
Problem: Imbalanced, limited dataset for robust model training.
Solution: Preprocessing, augmentation, advanced CNN architectures; GradCAM visualizations for clinical interpretability.
Stack: Python, TensorFlow, Keras, Medical Imaging, Data Augmentation
US Airline Sentiment Analysis [NLP, Finance, Customer Analytics] [GitHub]
Built a scalable NLP pipeline for airline Twitter sentiment analysis, enabling early detection of customer service pain points and driving social strategy improvements.
Impact: Helped reduce negative sentiment by 12% within three months of dashboard adoption.
Problem: Processing noisy, high-volume social data in real time.
Solution: Layered text cleaning, ML models (SVM, Random Forest), cross-validation for sustained accuracy above 88%.
Stack: Python, NLTK, scikit-learn, Pandas, Jupyter Notebook
PyTorch Capstone Project [Deep Learning, Computer Vision] [GitHub]
Developed modular CNN architectures for CIFAR-10 image classification, improving accuracy benchmarks by 4% and enabling rapid retraining for new domains.
Impact: Tools reused in capstone bootcamps, adopted for multiple academic initiatives.
Problem: Preventing overfitting and model bloat during deep learning.
Solution: Integrated advanced regularization (dropout, batch norm), extensive augmentation strategies.
Stack: Python, PyTorch, CNN, Computer Vision
CNN vs ANN Model Comparison [Machine Learning] [GitHub]
Compared CNN and ANN models for image classification, confirming CNNs’ superior accuracy and efficiency for vision tasks.
Problem: Selecting optimal models for practical deployment.
Solution: Benchmarked accuracy, training time, and resource use to guide future projects.
Stack: Python, TensorFlow, Keras, CIFAR-10
Snake Game [Game Development] [GitHub]
Created an interactive Snake game in Python with smooth GUI, responsive controls, and scalable architecture for teaching game logic.
Problem: Designing continuous real-time gameplay.
Solution: Used Pygame for modular event handling and dynamic scoring.
Stack: Python, Pygame
Web Scraping iPhone Reviews [Data Engineering] [GitHub]
Developed a dynamic scraper to extract real-time iPhone review and pricing data from Amazon, supporting competitor analytics and market trend analysis.
Problem: Rapidly changing HTML and anti-bot countermeasures.
Solution: Employed rotating user-agent, strategic delays, and error handling for robust automation.
Stack: Python, Requests, BeautifulSoup, Pandas
Eggplant Disease Diagnosis [Agriculture, Computer Vision] [GitHub]
Designed a transfer learning-based deep learning tool for eggplant disease classification, enabling actionable farmer interventions and reducing misdiagnosis by 30% in pilot trials.
Problem: Maximizing disease classification accuracy with limited samples.
Solution: Custom CNN architecture, field validation, and explainability integration.
Stack: Python, TensorFlow, CNN, Transfer Learning
Data Visualization Toolkit [Visualization, Data Science] [GitHub]
Developed reusable visualization tools for reporting and dashboarding, empowering non-technical teams to explore actionable insights from complex datasets.
Problem: Enabling fast, flexible analytics for all users.
Solution: Created modular APIs and templates using Matplotlib, Seaborn, Plotly.
Stack: Python, Matplotlib, Seaborn, Plotly, Pandas
Emotion AI [Computer Vision, HCI] [GitHub]
Implemented vision-based emotion recognition using CNNs and live feedback to support interactive sentiment analysis for chatbots and user research.
Problem: Detecting subtle, real-world emotions in variable conditions.
Solution: Integrated OpenCV and advanced neural architectures for real-time results.
Stack: Python, OpenCV, Keras, CNN, React, Flask