Varun Mathur Email & Phone Number

NLP developer working in the enhancement and development of applications based on Large Language Models (LLMs).Some of the projects I have worked on:Performance Evaluation of Rankers and RRF Techniques for Retrieval Pipelines:I have done a comparative study of adding different combinations of rankers in a Retrieval pipeline along with the use of Reciprocal Rank Fusion (RRF) techniques. The results were evaluated on four metrics: Normalized Discounted Cumulative Gain (NDCG), Mean Average Precision (MAP), Recall and Precision. We analyzed the effectiveness of adding different rankers to pipelines to improve the quality of retrieved documents.A comparison of Hyperparameter Tuning and Optimizer Selection on Training Efficiency and LLM Performance:Built a Question Answering system for News Articles. For the SQuAD dataset, a baseline was set using the BERT model. The BERT, DistilBERT and FinBERT models were fine-tuned using an AdamW optimizer.Built a sentiment analysis model to predict the sentiment of a Financial News article. For the Financial PhraseBank dataset, a baseline was set using the BERT model. The BERT, DistilBERT and RoBERTa models were fine-tuned using an AdamW optimizer.Built a summarization model that gives short and concise summaries for News Articles. For the Multi-News dataset, a baseline was set using the BART model. The BART, DistilBART models were fine-tuned using an AdamW optimizer.The optimizers were tuned for different parameters by a specific way of search spaces. Each parameter was first tuned on a large search space, and then a smaller and more refined search space was used to find the optimal value for training the model.A comparative study of different optimizers used for training was done. The optimizers were tuned for different parameters by a specific way of search spaces. Each parameter was first tuned on a large search space, and then a smaller and more refined search space was used to find the optimal value for training the model.Financial Dashboard : Built an end-to-end Financial Dashboard that collects and consolidates all of a business's critical observations in one place using the information obtained from the annual 10-K SEC Filings. It contains:- Insights and summaries for different sections from annual corporate filings.- Sentiment-based score that measures the company's performance over a certain time period.- Identification of Important topics and Frequently occurring words mentioned in the report.Email: varunm500@gmail.comGithub: https://github.com/vrunm