Akshita Sharma

Researched and analyzed numerous LLMs for translation accuracy across languages, finalizing the best-performing model for optimal product impact. Integrated the selected LLM in platform wrapped by a robust translation pipeline that enhances multilingual capabilities. Integrated and optimized open-source LLMs into the platform by doing R&D based on accuracy, efficiency, cost and latency. Designed and developed a Sentiment analysis service that integrates seamlessly with system, enabling users to analyze end user sentiments in real time. Spearheaded development of a dedicated for Bahasa language support, conducting in-depth research on models like paraphrase multilingual-mpnet-base-v2, all-mini-l12-v2, etc. Developed an AI feature that enables users to generate conversation summaries with end-users. Researched and developed the Generative AI Docusense feature based on LangChain, enabling users to process document efficiently. Developed the pipeline for Google Gemini, Generative AI Docusense allowing customers to process documents without migrating data to OpenAI.

Gathered extensive text data from Reddit, employing NLP libraries for data preprocessing. Conducted text classification analysis using diverse pre-trained and deep learning models, including BERT, Roberta, ALBERT, RNN, CNN, and BiLSTM. Explored various classical machine learning algorithms and Stacking to enhance the performance of the top-performing classifier, for classifying the sentiment on human blogs.