Soumendu Bhattacharjee, Ph.D Physics

Performed entity extraction from PubMed papers for cancer research using LLMs. Leveraged the multimodal capabilities of LLMs to enhance entity extraction performance. Addressed the challenge of maintaining 100% recall by designing a hybrid approach. Utilized BioBERT and SNOMED CT ontology-based methods to improve entity extraction accuracy. Researched and experimented with U-Net, Mask R-CNN, and Mask2Former for semantic, instance, and panoptic segmentation of satellite images. Developed an application for automated satellite image segmentation to support large-scale analysis.

Fine-tuned LLM models for classification tasks using Azure DevOps. Performed multi-label classification on large healthcare documents stored in Azure Blob using Spark. Used Azure Data Factory to automate preprocessing pipelines and summary generation workflows.

Provided supply chain optimization solutions. Integrated large language models (LLMs) with knowledge graphs for enterprise content navigation. Enhanced RAG system performance using graph-based approaches. Built Q&A systems and chatbots using LLMs. Generated reports for legal documents and news articles. Fine-tuned LLMs using LoRA and ran LLMs locally with Ollama. Built RAG Agent for tabular data Q&A and LangGraph research agent. Performed web scraping and data cleaning for LLM input. Led the data science team and addressed diverse business problems. Deployed code using Docker and FastAPI. Worked with MongoDB, Redis, Kafka, and Postgres for real-time content streaming. Built and deployed Streamlit apps using data from S3, Azure Blob, and SFTP. Managed operations and orchestrated jobs using Airflow and Cron scheduling. Performed speech-to-text analysis and A/B testing.

Used autoencoder models for fault detection in PV systems. Supervised team for long-term GHI prediction using multivariate LSTM, XGBoost, and CatBoost. Managed connector fault detection projects using Boosting, RFC, and SVM algorithms. Applied Fourier transform for noise reduction. Transformed tabular data into images and performed classification using MobileNet, CNN, VGG16, and VGG19. Generated synthetic time series data using Gretel and TimeGAN. Built classifiers by converting tabular data into image data and applying noise removal algorithms. Supervised junior colleagues and led multiple projects. Presented work to clients weekly. Trained and tested ML models on AWS SageMaker. Used AWS S3 and Glue for data ingestion and processing. Stored results in Snowflake and Redshift. Scheduled batch model evaluations using Airflow.

Performed classification tasks on tabular datasets using Random Forest, Boosting, LightGBM, and SVM. Conducted image classification with deep neural networks for over 1500 classes using InceptionV3, ResNet50, MobileNet, CNN, VGG16, VGG19, EfficientNetB7. Explored combining CNNs with Random Forest and XGBoost. Worked on image segmentation for biological images using UNet and Stardist. Conducted feature selection using PCA, Random Forest, and chi-square. Developed automated systems for classification from raw spectroscopic data. Analyzed tabular data by converting to images with Tab2Img. Performed cell and image segmentation using UNet, U2Net, and Stardist. Applied Fourier analysis for denoising time series. Supervised junior members and presented work to technical and non-technical audiences. Analyzed text data using Tf-idf, Word2vec, Glove, VADER, RNN, LSTM, GRU, BERT, GPT. Conducted time series analysis using ARIMA, SARIMA, Prophet, LSTM, Conv1D & Conv2D. Built pipelines integrating SQL Server, BigQuery, Duck DB. Used S3, MongoDB, and Redis for hybrid cloud-data storage.

Performed data analysis and optimized ML models for large-scale scientific experiments. Developed controls and electronics for detection systems. Prepared scientific proposals and documentation. Presented analysis results at conferences. Collaborated with global groups and supervised junior members.

Analyzed data and optimized ML models for scientific experiments. Published results in high-impact journals. Maintained and improved hardware and software equipment. Managed experimental campaigns and documentation. Presented results at conferences. Supervised undergraduate and master students. Collaborated with colleagues and group members.

Analyzed data and optimized ML models for scientific experiments. Published results in high-impact journals. Prepared scientific proposals and documentation. Presented results at conferences. Supervised undergraduate and master students and collaborated with other groups. Performed Monte Carlo simulations for gamma ray detectors using C++.