Inzamam Safi's Linkedin Analytics

--- Data scientist @subex --- As a data scientist at Subex, I played a key role in developing a NO code AI platform. I achieved significant milestones, including building a Name Entity Recognition model with an impressive 82.1% F1 score and creating a pipeline using Kubeflow. Additionally, I utilized the Dask library to parallelize machine learning tasks, reducing training time by 30% on large datasets. I also successfully implemented Dask-Optuna for automating hyperparameter optimization in parallel, resulting in a 23% time reduction. Furthermore, I contributed to annotation correction, improving model accuracy, and developed APIs using FastAPI for deep learning models. Lastly, I focused on dynamic resource allocation to optimize machine learning models based on data size and model requirements. --- M.Tech @ IIIT Allahabad in Data science --- During my Masters, I conducted a research project focused on extracting patient age from clinical trial data. Remarkably, we achieved a high F1 score of 72%, surpassing the performance of other works in the field. Notably, even the esteemed IBM Research Lab had a 10% lower score than our approach, highlighting the superiority of our methodology. Furthermore, I contributed to an image caption project where I implemented a hybrid approach to enhance model accuracy. Through the utilization of a knn model, I was able to achieve accuracy levels comparable to human performance. Impressively, our model attained a BLEU-4 score of 0.19, further validating the effectiveness of our approach in generating accurate image captions