Ayush Patel

Engineered a modular multi-agent system with specialized agents (General, AI, and Admissions) using dynamic routing to assign queries based on context. Implemented Q-learning algorithm that optimizes response parameters (temperature, top-p, max tokens) in real-time based on user feedback. Integrated LangChain's ConversationBufferMemory for multi-turn dialogue and FAISS vector store for past interaction retrieval. Augmented responses with Wikipedia API integration for factual data. Features a self-correcting feedback loop that penalizes generic answers and rewards high-quality outputs.

Developed a comprehensive suite of 3 dashboards (Summary, Overview, Details) tracking loan portfolio performance for a major financial institution. Engineered complex SQL queries and Power BI DAX measures for real-time KPI monitoring. Achieved clear visibility into 'Good' vs 'Bad' loan segmentation by DTI ratios, interest rates, and employment length. Implemented geospatial mapping and monthly trend forecasting to identify high-growth lending regions.

Developed a Convolutional Neural Network (CNN) to classify facial images into four emotional categories: Angry, Neutral, Bored, and Focused. Combined FER2013 dataset with 1,500+ scraped images. Implemented HAAR classifiers for face detection and Unsharp Masking for image enhancement. Achieved 70% accuracy using Adam optimizer with early stopping across 100 epochs.

Built a 10,000+ line-of-code game engine in Java utilizing MVC pattern and SOLID principles for modularity. Engineered 4 distinct AI player modes (Aggressive, Benevolent, Random, Cheater) using the Strategy Pattern. Achieved high system stability through JUnit 5 testing suites with automated tournament modes and map validation logic. Integrated Java Serialization for state management and a custom Map Editor for .map file parsing.