Swarna Mallik

Data Scientist with 4.5 years of experience in NLP and Deep Learning, and 2+ years in Generative AI. Key contributor to KnowRA+ - an award-winning, domain-customizable virtual assistant leveraging RAG, MiniLM, Qdrant, and more. Proven ability to deliver production-ready AI solutions that enhance service delivery. Skilled in Statistics, Machine Learning, Python, NLP, Hugging Face and Generative AI with a strong focus on continuous learning and innovation.

Ongoing (2023–2024)
Led the development of KnowRA+, an award-winning domain-adaptive virtual assistant built using Retrieval-Augmented Generation (RAG). Designed to handle domain-specific queries across travel, insurance, and more, it integrated MiniLM for semantic search and Qdrant for efficient vector storage, enabling scalable and context-aware responses.
Tech Stack: RAG, MiniLM, Qdrant, LangChain, Generative AI

Image processing and metadata extraction 

Aug 2021
Implemented OCR pipelines for processing highly unstructured healthcare documents, enabling accurate extraction of
critical information from scanned records and handwritten text. Also applied image preprocessing and deep learning techniques to
improve data quality and downstream NLP performance in clinical and administrative workflows.
• Tech stack: OCR, Computer vision, OpenCV, NLP techniques

Data Anonymisation 

Aug 2024
Developed an advanced anonymization system utilizing Named Entity Recognition (NER) to process unstructured legal
text documents. The system effectively transforms PII and other sensitive data to minimize the risk of re-identification,
ensuring strict compliance with data privacy regulations and significantly enhancing information security.
• Tech stack: ML, NLP, faker etc.

Lumbar Spine Degenerative Classification 

Aug 2024
Built a comprehensive classification system for lumbar spine degeneration using Magnetic Resonance Imaging (MRI)
scans, aimed at detecting and enhancing diagnostic accuracy and clinical decision-making.
• Tech stack: Various neural network models like EfficientNet, InceptionV3, ResNet etc.