Hire Nazira S. - Generative AI Architect

* Gen AI - * Conducted PoCs and benchmarked LLMs (GPT Based Models, LLaMA all versions, Alpaca, RoBERTa, * Gemma, Mistral, etc.) to evaluate performance on domain-specific data. * ICD and CPT code Prediction: Model-driven approach to predict (Disease, Organ, Impact) and performed fast text search for ICD code retrieval from AIPC data * Denials Classification: LLM-based Denial Reason Extraction and FAISS-powered embedding search to predict Denial Codes * Automated Entity Extraction: Reading metadata from DB, extracting entities using LLM, post process and update the DB * Table Extraction: LayoutLMv3 for structured data retrieval and analysis. * Document Text Extraction: DocTR, Tesseract OCR, Nuance - Extracting texts from PDF and TIFF format * Predictive AI- * Attrition Analysis: Predict resource attritions using ML models. * LLM Engineering & Deployment- * Implemented RAG-based AI solutions leveraging LlamaIndex and LangChain for enhanced knowledge retrieval. * Designed and deployed end-to-end ML pipelines, integrating FastAPI with fine-tuned models. * Optimized CPU/GPU consumption, leveraging multi-threading, parallel processing, and vectorized operations. * Integrated SQL databases for structured data processing in AI workflows. * Deployed open-source LLMs (LLaMA, Mistral, Gemma) from Hugging Face, avoiding reliance on OpenAI APIs. * Image Preprocessing & OCR- * Open CV

* Roles and Responsibilities: * Performed multiple approaches like weak supervision, unsupervised data augmentation to obtain high performance and low latency classification model for documents * Built training pipeline and prediction pipeline for classifier to train with different datasets * Image to Text conversion using image pre-processing techniques and custom training of Tesseract to obtain high precision text data extraction from documents * Entity extraction of extractive and abstractive fields using Q&A models and language models and validation using distance supervision, NER and Regex to obtain high confidence fields of interest * Built a knowledge graph to represent knowledge obtained from documents ML/AI Solutions: Smart-Ops - Building Intelligent Process Automation Platform having capabilities on Vision, Speech and process mining.