SHAILENDRA UPADHYAY

Designed, trained, and operationalized a comprehensive suite of 250+ SpaCY-based Named Entity Recognition (NER) models, adeptly utilized for precision metadata extraction in diverse applications. Developed and deployed web applications using frameworks such as Flask, Django, FastAPI, and Streamlit to showcase AI capabilities and deliver user- friendly interfaces. Designed and implemented robust data pipelines to streamline information flow and optimize data processing efficiency.

Proficiently fine-tuned and operationalized Language Model Architectures (LLMs), including BERT and FALCON-7B, specializing in metadata extraction and classification functionalities. Extracting metadata using RAG-based techniques with LLMs such as GPT4 and Zypher-7B, leveraging contextual information from a vector DB through tailored prompts and specific instructions. Developed and deployed conversational chatbots and AI models that generate natural language responses and understand user intents. Deployed a zero-scale Kubernetes solution for optimal resource consumption according to the traffic on models for prediction. Conducted data analysis, modeling, and statistical inference to derive meaningful insights and improve AI model performance. Semantic search based application using LLMs for embeddings and pinecone for vector database.

Implemented functionality utilizing LayoutLM, a multi-modal language model, for title extraction, while also leveraging other transformer-based models like the T-5 model for summarization purposes. Implemented Docker for deploying models on Torchserve to facilitate the deployment, serving, and scaling of PyTorch models in a production environment. Fine-tuned and deployed the YOLO V-5 model to pinpoint signature coordinates within legal contracts, facilitating the extraction of signature box coordinates and associated metadata fields.