SATYAJIT SAHOO

Sr. Decision Scientist

Mantrana Strategies Pvt Ltd

01.2023 - Current

Analyzed survey and prediction errors for the 2024 General Elections using Python, Machine Learning (XGBoost), SHAP values, and VIF; identified key KPIs driving survey errors, uncovered sampling biases, and improved prediction accuracy by leveraging structured data from past surveys and the Election Commission of India.
Designed and executed the “Mood of the Nation” survey, reducing operational costs by 25% through a mixed-mode (CAPI, CATI, CAWI) approach while maintaining sample integrity; applied Forest Plots, ANOVA, Chi-Squared tests, and Power Analysis to validate findings and ensure robust statistical accuracy.
Engineered and maintained dynamic dashboards in Google Sheets to monitor survey performance across multiple administrative levels; integrated data from CAPI, CATI, OS, and historical election results, automated 10-hour refresh cycles, and displayed key KPI metrics for real-time situational analysis and decision-making.
Automated voter and candidate data collection pipelines by web scraping electoral rolls (ECI) across all Assembly Constituencies and candidate nominations using Selenium; applied OCR to extract EPIC IDs, built a Random Forest model to solve CAPTCHAs, and scraped voter details into structured JSON datasets. Extended scraping workflows for UP Local Body Elections and Lok Sabha 2024 candidate nominations, scheduled periodic runs every 4 hours, and integrated outputs into dynamic dashboards to track ticket distribution and electoral insights in real time.
Built ETL pipelines to extract large-scale electoral survey data from MySQL and Google Sheets, transform it using Python (Pandas, NumPy) for cleaning, standardization, and validation, and load outputs back into Google Sheets; integrated with Cron for automated scheduling and monitoring. Additionally, engineered ETL workflows to standardize survey data via Google Sheets and automate PDF report generation using WeasyPrint, streamlining reporting and analysis.
Applied normalization techniques and k-means clustering on historical election data to categorize villages for strategic party planning, enabling a data-driven understanding of electoral landscapes and improving precision in targeting and grassroots engagement strategies.
Implemented Iterative Proportional Fitting (IPF) on survey data to align samples with population proportions, ensuring robust normalization; conducted detailed analyses across multiple administrative levels to uncover nuanced political scenarios and provide a comprehensive understanding of the electoral landscape.
Engineered a candidate selection model using logistic regression on survey data and historical voting trends, analyzing key factors influencing candidate success and enhancing the accuracy and precision of candidate selection decisions.