Sakshi Priya Email and Phone Number

Trained a llm model for multi class classification problem and achieved 97% accuracy.Currently working on optimising the virtual agent model to reduce its latency.

Fine-tuned the NLLB model to enhance multilingual translation accuracy. Further implemented optimization techniques like model quantization to reduce latency and deployed the optimized model using FastAPI.Prepared datasets for hackathon organized at the Alan Turing Institute by Kunato. It involved web scraping, model building, and discussions with the institute. Added a new feature to the product that displays activation maps for pricing models using SHAP and LIME, enhancing explainability for end users.Worked on the development and deployment of content genome models aimed at predicting the quality and time relevance of articles. Utilized large language models (LLMs) such as LLaMA3 to augment datasets and applied BERT Multilingual for quality classification tasks. For the time relevance model, used zero-shot BERTopic to find the topics and based on number of peaks determined article longevity.Fine tuned large language model llama 3.1 by utilizing PEFT technique for identifying the clarity of article.

Improved the accuracy of the string matching algorithm by 7% for identifying healthcare providers' names on websites. Addressing challenges related to different name orders, various arrangements, and truncated forms of names.Implemented an efficient Named Entity Recognition (NER) model to extract provider names from webpages. Leveraged medical credentials strategically to narrow down the search space, optimizing the algorithm's efficiency.Implemented FPG and Bayesian modeling techniques to identify associations within claims and clinical data, resulting in an increased member population receiving care and a 15% improvement in the HEDIS score for a health plan.

Implemented forecasting models like ARIMA, SARIMA, and exponential smoothing to predict member attribution using historical data. This initiative led to the identification of member churn in real-time, resulting in substantial annual savings of approximately $7.7M for a healthplan.Spearheaded the development of an end-to-end pipeline capable of ingesting and transforming data from 55 diverse sources. This comprehensive system processed around 1 billion records daily. Additionally, automated daily alerts to stakeholders to ensure proactive issue management.

Designed the GUI framework for cancer drug database. Built a machine learning model based on Breast Cancer data and used web crawling for data exploration. The project aims to observe helpful features in predicting malignant or benign cancer and to see general trends that may aid in model selection and hyperparameter selection.

Explored the Southeast Asian Hydroponic market for expanding the business. Monitored competitor's marketing activities and prepared regular sales forecasting reports.

Designed algorithm to identify all the identical repeats in both protein and nucleotide sequences.The algorithm developed was unique in its utilization of dynamic programming rather than local alignment and its speed compared to other algorithms. Furthermore, the algorithm has been implemented as a web-based computing engine and database to give the user precise and clear output.