Swarnabha Roy

About Swarnabha Roy

Data Scientist with 2+ years of experience building production AI systems that deliver measurable business impact.At Elutions, I work across the full ML lifecycle, developing anomaly detection systems for industrial operations and RAG-powered AI assistants using LangChain and agentic workflows. My focus is on productionizing AI solutions with robust MLOps practices, from experiment tracking to scalable deployment.I am passionate about GenAI/LLM applications, agentic AI architectures, and building reliable ML systems at scale.Always happy to connect with fellow builders working on interesting problems. Let\'s chat!

Experience

1. Data Scientist

   Elutions

   Jun 2023 — Present · Tampa, FL, US

   Built internal RAG system using LangChain, ChromaDB, and agentic workflows for natural language querying via semantic search and tool-calling (SQL, Python), saving 16+ hours of manual effort per employee monthly• Developed end-to-end industrial anomaly detection using PyTorch Autoencoder, LSTM, and Autoformer architectures for a manufacturing client, monitoring 50+ sensor streams at 5-min resolution with 85% recall, 60% precision, and 6–12 hour advance failure prediction• Architected predictive maintenance pipeline for petrochemical operations using PatchTST and TiDE models on 290K+ multivariate readings from 7 cracking furnaces, with automated alerts, lead-time quantification, and diagnostic visualizations• Designed a 4-component confidence scoring framework (feature quality, temporal consistency, model uncertainty) with adaptive debouncing algorithms, reducing false alarms by 40% while maintaining detection rate• Led cross-functional optimization for a water desalination plant through EDA, feature engineering, XGBoost/ensemble modeling, and constraint optimization across 5 KPIs, delivering $13.6M in client benefits via 11% resource efficiency and 7% yield gains• Engineered MLOps transformation of Maestro AI by automating training pipelines with MLflow, ZenML, and Git-based version control, reducing deployment time by 60%• Implemented causal inference and A/B testing frameworks with statistical hypothesis testing to validate operational interventions and drive data-driven decisions

Education

• University of Florida

  Master of Science - MS, Computer and Information Sciences

• University of West London

  Grade 8, Electronic Keyboard

• ABRSM

  Grade 4, Music Theory and Composition

• University of West London

  Grade 6, Music Theory and Composition

• Institute Of Engineering and Management

  Bachelor of Technology, Computer Science & Engineering

  2017 — 2021