● Design and develop service-based GAI Bot enterprise application, collaborated with multiple teams so that same application can be used by multiple projects without standalone deployment. ● Implemented Retrieval Augmented Generation (RAG) model, that leverages already trained client specific documents to augment its knowledge and achieve state of the art results. ● We have used Haystack pipelines for content segmentation, schema creation, document retrieval, GPT-4 openAI generative answer generation and query prompt engineering. ● We have also use multihop multi-conversation chatbot, enabling users to interactively retrieve search results through natural language conversations. ● Designed Unstructured document processing pipeline by first extracting important section sentences from the documents using multi-qa-distilbert, k-means clustering, and GPT-4 azure openAI.● Worked on Event Detection pipeline, use zero-shot learning facebook/bart-large model approach to create data for business related events from the business documents and news. ● Developed and deploy on-premises generative model, which is alternative to open AI and integrated LLM model with GPU supported code, deploy on GPU enable multi worker cluster. ● Create parsing pipeline to automatic extracting text from excel, pdf, doc, ppt and preprocessing and merge the text for further business requirements. ● Develop on-boarding code, data synch and query pipeline to fully integrate with client project. ● Use matrix-based score saving method for each record, turnaround time for recommendation engine ● Build and tested own word Embedding vector space model using Fasttext model and did fine tuning using Sklearn Grid search algorithm with different parameters. ● Effectively implement pretrained model to convert text into high dimensional vectors then use cosine similarity to find out the Semantic Similarity between past documents, with user feedback mechanism.

*Use advanced BERT base model uses 12 layers of transformer encoders for Emotion Analysis and classify text into HAPPY, SAD, LOVE, ANGRY and FEAR labels.*Study different types of model available for implementing our Natural language processing task (FastText, CNN, LSTM, SVM classifier, logistic Regression) and implement them according to our needs. *Design a software architecture for Sentiment Analysis on 1.6 million twitter tweets including emojis using deep learning framework Keras on the backend tf2.0.*Use Google Colab to train the deep learning model which can CNN-LSTM pre-trained models to classify sentiments POSITIVE, NEGATIVE and NEUTRAL.*Create multiple MySQL database for the user-login, token, subscription and rate-limiter based architecture using python Object Relational Mapper (ORM) sqlalchemy.*For text analysis on the web browser, develop and design a RESTful API for the client to find out the emotion of the text or tweet using FLASK web application framework endpoints.

*Used Beautiful-soup module for web scraping and web page crawling for extraction of data from pappy joy Data management system, patient profiling for effectiveness and better future planning. *Build a software using data mining and regex module for transfer thousands of documented files entry into the excel sheet and extracted important data from document files. *Increased the proficiency of the tinnitus relief device by increase the variant of the different type of devices. *Extracted key problem faced by the individual patient from the data base using NLP(Natural language processing), word tags count and TF-IDF(Term frequency-inverse document frequenvy)