Atharva Mehul Patil
Projects
CourseAid - A better RateMyProfessors
Course selection is overwhelming for master's students. Existing platforms like Rate My Professors focus on instructors, not courses holistically. Our centralized hub integrates course descriptions, student reviews, professor-based difficulty ratings, and robust search/filtering to help students make informed academic decisions.
FlexiCal
Desktop calendar application supporting multiple calendars across different timezones, recurring events with flexible editing patterns, and export functionality compatible with Google Calendar and Apple Calendar. Delivered through iterative development with both command-line and graphical interfaces.
BookSphere - Contactless Library Access & Management System
BookSphere is a comprehensive Android-based application developed for university use. The key features include distinct interfaces for students and administrators, QR code integration for seamless physical entry, and efficient management of student data, book information, and entry records through Firebase Realtime Database and Authentication.
Research
Genetic Algorithm-Driven Feature Selection Optimization for Skin Cancer Classification
9th International Conference on Soft Computing: Theories and Application (Springer), 2024
Addresses the challenge of optimizing feature selection in CNNs for skin cancer classification (benign vs. malignant). Uses VGG-16 for feature extraction, applies PCA combined with Genetic Algorithm (GA) for feature optimization, and compares classification performance using Artificial Neural Network and Random Forest classifiers. The GA-driven PCA approach achieves higher accuracy with 50% fewer features compared to PCA alone.
Unmasking Hallucinations in LLMs Using Analysis of the LLAMA 2 Model and RAG Intervention
2nd International Conference on Advances in Technology and Management (STM Journals), 2024
Tackles the hallucination problem in financial trading chatbots where AI generates false or unverifiable information. Implements "TradeBot" using Llama 2 Model with Retrieval Augmented Generation (RAG) that references the NCFM book as an external knowledge source to ground responses in verified financial information. Results demonstrate that RAG significantly reduces hallucinations and improves response reliability compared to the base Llama 2 model without RAG.
Internship Experience
January 2025 - June 2025
Quant Intern | ORIM Advisors Pvt Ltd
Developed ML pipeline automating financial research for 5K+ stocks using peer clustering and open-source LLMs, significantly reducing inference costs. Built causal inference engine linking news sentiment to stock movements, quantifying market psychology indicators
May 2023 - August 2023
Industrial Training (DevOps Intern) | Datamato Technologies Pvt Ltd
Engineered CI/CD pipeline from scratch using GitLab, Docker, and YAML for automated deployment of Node.js employee portal. Implemented testing automation including lint testing (Flake8) and unit/smoke testing (Pytest)