Varalakshmi Garidapuri

Developed and supported AI/ML and Generative AI solutions using Python, PySpark, Azure Databricks, Delta Lake, and MLflow across structured and unstructured data. Built reusable data cleansing/transformation/feature engineering workflows, performed EDA and data-quality checks, and evaluated ML workflows with appropriate validation and metrics. Implemented NLP and LLM pipelines for extracting topics, intents, entities, sentiment, and themes, and designed end-to-end RAG systems with Azure OpenAI, LangChain, Azure AI Search, embeddings, and semantic retrieval (including ingestion, OCR/text extraction, chunking, metadata enrichment, retrieval, prompt construction, and response generation). Created evaluation frameworks, ran controlled experiments for prompt/retrieval configurations, and conducted structured error analysis for retrieval failures and factual inconsistencies. Built FastAPI services for inference and semantic search, containerized with Docker, supported CI/CD via Git/GitHub Ac

Built Python-based NLP and ML workflows to process large collections of PDFs and scanned/unstructured documents. Developed OCR pipelines using Tesseract and AWS Textract, created preprocessing utilities to clean/normalize extracted text, and supported multilingual processing including English and Telugu. Used spaCy, NLTK, and Hugging Face Transformers for document classification, text similarity, and information extraction, including embedding generation for similarity and downstream analytics. Designed classification/categorization workflows and NLP analytics to extract recurring subjects and key entities, implemented data-quality checks across OCR and preprocessing, and ran controlled experiments comparing OCR/preprocessing and transformer approaches with representative evaluation datasets. Performed structured error analysis (OCR vs preprocessing vs model errors), reported quantitative findings, and automated OCR/NLP pipelines to reduce manual effort.

Delivered AI/NLP and machine learning solutions for healthcare customer-service operations using call recordings and conversation transcripts. Built speech analytics pipelines converting audio to text using Mozilla DeepSpeech, then developed transcription cleanup and normalization to address recognition noise, punctuation issues, and inconsistent formatting. Developed intent classification models for customer needs/topics and sentiment analysis for quality monitoring and operational analytics. Prepared labeled training/validation datasets with cleansing, feature preparation, and quality checks, conducted controlled experiments across preprocessing/feature engineering and modeling choices, and performed structured error analysis for misclassifications and weak representations. Communicated results through Power BI and Tableau dashboards and improved production workflow reliability with logging, validation, error handling, and monitoring.