Sai Subhransu Dash

I’m a Data Scientist and AIML Engineer focused on delivering data-driven solutions in both classical ML and modern LLM applications. I’ve worked on ML-based prediction systems and LLM-powered multi-agent pipelines with the goal of making them production-ready.

As an AI / ML Intern at VG Marine Tech Pvt. Ltd., I developed an AI/ML-based Fault Detection and Early Warning System for EV battery and traction motor telemetry. Using physics-based pseudo-labeling and unsupervised anomaly detection (Isolation Forest), I transitioned to supervised models with XGBoost (91.27%) and RandomForest (91.21%), exporting deployable pickle files for integration into live dashboards.

I also engineered 26 statistical and frequency-domain features for a BLDC fault dataset (184 samples) and achieved 100% test accuracy with Decision Tree and Random Forest. Then I reduced the feature set to the top 5 predictors, retaining 97.3% (Random Forest) and 94.6% (XGBoost) accuracy, while serializing models via joblib for deployment.

In my projects, I built a Multi-Agent AI Research System using LangChain with Groq, Tavily, and Streamlit—creating a Search Agent and Reader Agent plus LCEL Writer and Critic chains for structured report generation and quality scoring. I’ve also built a 3-class BMS Motor Health Classifier (Normal / Warning / Critical) on 50,000 CAN bus samples using physics-based features and probability-based early warning logic to flag critical faults proactively.