Data Scientist
Sanidhya Karnik
Building ML models, data pipelines, and AI-powered solutions. Turning complex data into actionable insights.
Data Scientist
Building ML models, data pipelines, and AI-powered solutions. Turning complex data into actionable insights.
// About
I'm a Data Scientist with 3.5+ years of experience building machine learning models, forecasting systems, and analytics solutions across finance, hospitality, and tech.
I hold an MS in Data Analytics Engineering from Northeastern University and a BTech in Mechanical Engineering from IIT Madras. My work spans LLM applications, time-series forecasting, pricing optimization, and conversion analytics.
Currently exploring opportunities in data science and ML engineering where I can build systems that drive real business impact.
// Work
01
An autonomous multi-agent system for data pipeline monitoring and repair using Amazon Nova 2 Lite. Features four specialized agents (Monitor, Diagnostics, Repair, Orchestrator) with real dbt model integration, FastAPI backend, React dashboard, and full AWS deployment with EC2 and RDS PostgreSQL.
02
A production-ready Model Context Protocol server for secure LLM interactions. Features token-based authentication, role-based access control (RBAC), PostgreSQL audit trails, and multi-API integrations including Tavily search and weather APIs. Containerized with Docker for easy deployment.
03
End-to-end MLOps platform for predicting 30-day hospital readmissions using MIMIC-IV clinical data. Features XGBoost model (AUC: 0.72) with SHAP explainability, dbt data transformations, FastAPI serving with real-time explanations, Airflow orchestration, and production-ready Kubernetes/Terraform infrastructure.
04
An LLM-powered course recommendation system using Agentic RAG architecture. Combines content-based and collaborative filtering with real-time web search to deliver personalized learning paths for students transitioning into data science.
05
Fine-tuned transformer models (RoBERTa, BERT, DeBERTa) on the VUA Metaphor Corpus and multi-domain data from Reddit, IMDb, and news sources. Implemented class-weighted loss functions to handle imbalanced datasets for metaphor and irony detection.
// Toolkit