Lakshya Bhardwaj

Implemented and optimized recommendation systems for IndiaMART's sellers and buyers, resulting in a substantial 40% increase in purchases facilitated by the recommendation widget. Applied the innovative "Harry Potter Solution" at both seller and category levels, revolutionizing recommendation strategies for enhanced, tailored suggestions over generic approaches. Developed a seller-centric system aimed at boosting leads through recommended search titles, enhancing user engagement and seller satisfaction. Spearheaded the creation of A/B testing frameworks, enabling systematic assessment of system changes for informed decision-making and seamless integration. Extracted and organized medicine-related data, enhancing accessibility and search capabilities in ChromaDB. Implemented a sophisticated chatbot utilizing Llama2, empowered by the RAG (Retrieval-Augmented Generation) pipeline for improved user interactions. Led the fine-tuning of the llama2 7B chat model to accurately map diverse products to optimal categories. Executed rigorous data preprocessing and formatting to ensure seamless compatibility with Language Model (LLM) standards. Demonstrated analytical proficiency by achieving an impressive 82% accuracy through strategic model refinement using PEFT & SFT methodologies.

Created a Neo4j graph database to efficiently store and manage the MCAT-seller relationship, utilizing graph-based data representation. Devised a solution to enhance keyword search accuracy within product specifications, resulting in a notable 15% increase in precision.

Utilized the dlib library for Face Recognition in a live feed and optimized the algorithm by replacing the Neural Network for face detection with HaarCascade. Created a cricket regression model to give customers realistic betting options. Utilized the Mediapipe model to predict cricket umpire signals in real-time using landmarks' calculations.