Open to opportunities

Anubhav De

@anubhavde

Message

I build machine learning and LLM pipelines that turn complex data into usable insights.

United States

Message

What I'm looking for

I'm looking to apply machine learning, LLMs, and data engineering to production products, especially where reliable AI systems can reduce manual work and solve complex real-world problems.

At ZeroNorth, I engineered an automated insights extraction pipeline using Gemini and Mistral that saved the engineering team hundreds of hours per month. I deployed it on AWS ECS with Docker and CI/CD automation to support real-time vessel-data ingestion across the SMARTSHIP platform.

I also designed a Hugging Face and LangChain chatbot that handled hundreds of monthly user queries and reduced average report-processing time to under five minutes.

Previously, I researched Physics-Informed Neural Networks at National Chin-Yi University of Technology and built computer-vision models at VIT, including a CNN and U-Net histology segmentation model that reached 84.17% accuracy. I hold an MSE in Computer Science from Johns Hopkins University and have published peer-reviewed research on interpretable machine learning for clinicians.

Experience

Work history, roles, and key accomplishments

Data Science Intern

ZeroNorth

Jul 2025 - Nov 2025 (4 months)

- Engineered an automated insights extraction pipeline experimenting with multiple LLMs (Gemini, Mistral), to reduce manual data entry time and saved engineering team hundreds of hours per month.

- Deployed the pipeline to AWS ECS with Docker and CI/CD automation, ensuring high availability and enabling real-time data ingestion for vessel data across the SMARTSHIP platform.

- Designed and integr

LLM Gemini Mistral AWS ECS Docker CI CD Hugging Face Unit Testing

Machine Learning Research Intern

National Chin-Yi University of Technology

Sep 2023 - Mar 2024 (6 months)

- Researched and implemented Physics Informed Neural Networks (PINNs) with NVIDIA Modulus, DeepXDE, PyTorch, and TensorFlow, achieving lower error rates compared to conventional deep learning models on fluid–solid heat transfer problems.

- Modeled on 5 complex PDMS geometries with microchannels, showing potential to accelerate CAD-based simulations by decreasing computational overhead in solving

Physics Informed Neural Networks NumPy Machine Learning Deep Learning Machine Learning Algorithms Responsible AI Algorithms NVIDIA Modulus DeepXDE PyTorch TensorFlow Tensorboard Git

Undergraduate Researcher (Machine Learning)

Vellore Institute of Technology (VIT)

Apr 2023 - Mar 2024 (11 months)

- First Paper: Developed a hybrid CNN + U-Net segmentation model achieved 84.17% accuracy in classifying histology images, outperforming existing baselines by double-digit margins.

- Second Paper: Enhanced skin disease classification through fine-tuning of pre-trained models (ResNet, VGG, EfficientNet), yielding notable improvements in predictive accuracy.

- Published results in a peer-reviewed