Headline

“Data Scientist & ML Engineer | MLOps · NLP · Recommender Systems | PyTorch · FastAPI · Docker | Open to Work”

About

Data Scientist and ML Engineer building production-grade ML systems end-to-end — from raw data to deployed APIs. I specialize in MLOps pipelines, NLP with Transformers, and recommender systems. My projects process 2.75M+ records, are tracked with MLflow, versioned with DVC, containerized with Docker, and deployed on Render and Hugging Face Spaces. Key achievements: • Reduced Docker image from 11.5 GB → 434 MB enabling cloud deployment • Improved F1 score by 65.5% fine-tuning Weighted RoBERTa on LIAR dataset • Built a 7-stage reproducible MLOps pipeline with CI/CD via GitHub Actions Currently learning: RAG (Retrieval-Augmented Generation) and LLM fine-tuning. Open to Data Scientist, ML Engineer, and AI/ML roles. 📧 beherasuman44444@gmail.com | 🌐 sSuman Behera | Data Scientist Portfolio

Location

Bhubaneswar

Work experience

Machine Learning Engineer — Portfolio Projects at Self Employed, Jan 2025 - Present · 1 yr 6 mos, Bhubaneswar, Odisha, India, • Built a 7-stage Customer Churn MLOps Pipeline (SMOTEENN + GradientBoosting) achieving ROC-AUC 0.837, Recall 0.813 — deployed as FastAPI on Render via Docker with GitHub Actions CI/CD • Engineered a Hybrid E-Commerce Recommender on 2.75M+ interactions (Item-CF + NCF + GRU4Rec); reduced Docker image from 11.5 GB → 434 MB enabling production deployment • Fine-tuned Weighted RoBERTa for fake news detection — F1 improved 65.5% (0.3423 → 0.5664) over TF-IDF baseline; benchmarked 5 models with MLflow; deployed on Hugging Face Spaces • Tools used across all projects: PyTorch, Scikit-learn, FastAPI, Docker, MLflow, DVC, GitHub Actions, Render, Hugging Face

Education

Master of Computer Applications, Computer Programming, Specific Applications, Gandhi Institute of Engineering and Technology (GIET)

Bachelor of Science, PHYSICAL SCIENCES, Divine Degree College, Nayagarh