May 2025 – Present · 1 yr 1 mo

May 2023 – May 2025 · 2 yrs

• Spearheaded a high-performing Machine Learning team in the development of a cutting-edge HR-focused large language model. Achieved this by conducting extensive fine-tuning of the Llama 2 model on a colossal dataset consisting of 50 billion HR-related tokens.
• Demonstrated expertise in optimizing Large Language Models (LLMs) for specific applications through the implementation of parameter-efficient methodologies such as Lora and QLORA.
• Pioneered the efficient deployment of Large Language Models (LLMs) by applying state-of-the-art quantization techniques like GPTQ and GGML, along with leveraging optimization tools like VLLM.
• Innovated the creation of a dynamic Q&A tool for our support team. This transformative tool harnessed the power of vector databases, utilizing past support tickets to provide context-aware responses through the implementation of the RAG (Retrieval-Augmented Generation) model.
• Elevated the performance of our widgets by designing and training Learning-to-Rank (LTR) models, resulting in significant enhancements in NDCG (Normalized Discounted Cumulative Gain) scores.

Apr 2021 – May 2023 · 2 yrs 1 mo

• Pioneered the development and fine-tuning of a domain-specific BERT model utilizing contextual data. This breakthrough led to an outstanding 5-8% enhancement in the accuracy of downstream models.
• Utilized PyTorch BigGraph to architect and train an innovative graph embedding model, revolutionizing the prediction of crucial missing features for candidate profiles. The resultant model significantly heightened the precision and relevance of candidate recommendations.
• Engineered and implemented hierarchical classifiers capable of handling vast datasets with thousands of classes. This innovation laid the foundation for an Onet occupation classifier, categorizing jobs into 1110 distinct Onet occupations. The resulting augmentation of recommendation engine performance by 5% validated the impact and efficiency of the classification models.
• Devised and trained an ordinal classification model to accurately forecast both the experience range for a job and its corresponding Onet job zone.This innovation not only enhanced candidate-job matching by aligning experience ranges but also optimized the elimination of irrelevant results through precise Onet job zone matching.

Dec 2019 – Apr 2021 · 1 yr 4 mos

• Engineered a cutting-edge recommendation algorithm that delivers highly personalized job recommendations to candidates, leveraging their unique profiles. Achieved a remarkable 25% increase in the click-through rate, underscoring the algorithm's effectiveness in engaging users and driving user interaction.
• Spearheaded the development of a robust ensemble model capable of discerning and ranking the significance of skills within job descriptions, candidate profiles, and resumes. This innovative model streamlines the hiring process by providing clear insights into the most critical skills required for a role, enhancing the overall quality of candidate assessments.