Neharika Mazumdar

• Industrial Automation via OCR: Built end-to-end automation using DenseNet and PaddleOCR to extract tabular data from PDF diagrams and map it to real-time screen elements. Enabled automated actions within client tools, reducing manual mapping by 60–70%.
• Multimodal RAG System: Built a pipeline using CLIP, ALIGN, and ColPali to generate joint embeddings from PDFs (text, images, tables). Integrated with Milvus and LLMs (OpenAI, Claude) to enable multimodal Q&A, achieving ~20–30% improvement in semantic answer accuracy through embedding comparisons and prompt tuning.
• LLM Ops Integration: Evaluated Arize AI, Azure ML, and Evidently for monitoring drift, bias, and performance issues across deployed ML/LLM models; set up monitoring pipelines and alerts.
• Toxicity Detection Tool: Designed a content moderation pipeline using BERT, DistilBERT, OpenAI GPT, and Google Perspective API; conducted comparative analysis of models across model hubs.
• Data Engineering Projects: Built real-time data ingestion and aggregation pipelines using Apache Spark, Kafka, and Apache Flink.

Tech Stack: Python, Spark, Kafka, Azure ML, OpenAI, CLIP, ALIGN, HuggingFace, Google APIs, REST, PyTorch, TensorFlow, Flask

• Developed unsupervised clustering models (e.g., agglomerative clustering) to analyze semantic similarity and groupings across standardized dental codes, aiding taxonomy refinement.
• Built fraud detection pipelines over large-scale dental insurance claims using scikit-learn classifiers, with engineered features from cleaned and normalized fields (names, addresses, codes).
• Conducted EDA, feature engineering, and model evaluation to support fraud analytics and anomaly detection workflows.

Tech Stack: Python, scikit-learn, pandas, NumPy, SciPy, matplotlib, Jupyter, Regex.

• Applied ML and graph algorithms to biomedical and social data for temporal modeling and network analysis.
• Modeled viral transmission using MST-based seriation and dimensionality reduction (PCA, MDS) on high-dimensional genetic data.
• Built HCV host-pathogen graphs to trace transmission pathways.
• Analyzed Reddit opioid user behavior using NLP and social network graphs; applied SVM and Random Forest for classification.
• Developed an interactive web tool to visualize seriation and temporal progression in biomedical datasets.

Tech Stack: Python, BioPython, NetworkX, scikit-learn, Flask, PCA, MSTs, TF-IDF, Plotly

• Seagate Big Data Platform: Built batch processing jobs and managed multi-cluster Hadoop environments.
• Cisco SMA: Built REST APIs and implemented NLP pipelines for analyzing product-related Twitter data.
• Cloud Security Testing: Developed API testing tools and automation for security apps across AWS, Box, and Salesforce.

Tech Stack: Java, Python, Hadoop, Hive, Sqoop, Spark, Cassandra, Tableau, REST, NLTK, Selenium, TestNG