Mandira Sawkar

I am a Master’s student in Artificial Intelligence at Rochester Institute of Technology, actively seeking AI roles. My interests lie in natural language processing, explainable AI, and socially aligned machine learning.

I have pursued several applied and research-oriented projects as well as published a paper at EMNLP. I’m also currently pursuing thesis research on gender bias in the UK parliament during the suffrage movement. My background also includes experience in large-scale data analysis, model development, and system deployment.

Education

Rochester Institute of Technology, NY, USA, 2026
- Pursuing MS in Artificial Intelligence
- Coursework: Fundamentals of AI/ML, Math Methods for AI, Ethics of AI, Forecasting Methods, Deep Learning.
- Thesis: The Sound of Suffrage - Modeling Gender and Power in Parliamentary Speech.
IIIT – Hyderabad, Hyderabad, India, 2023
- Secured a Post Graduate Certification in Software Engineering for Data Science with Distinction.
- Coursework: Data Engineering, Machine Learning, Intelligent Systems Design, etc.
Indira Gandhi National Open University, Delhi, India, 2020
- Graduated with a Bachelor of Computer Applications with First Class.
- Coursework: Data Structures and Algorithms, Discrete Mathematics, Operating Systems, etc.

Research Interests

I’m broadly interested in building AI systems that are socially aligned and interpretable, especially in high-stakes domains like justice and governance. My current focus areas include:

Socially Aligned AI – Designing frameworks that integrate fairness, optimization, and behavioral modeling to reduce systemic bias.
Natural Language Processing & Reasoning with LLMs – Exploring how large language models can emulate nuanced human reasoning and decision-making.
Multilingual LLMs & AI Translation – Developing inclusive AI systems that support equitable access across languages, and consequently across different demographics.
Explainable AI – Creating interpretable models to increase transparency and stakeholder trust in machine learning systems.

Professional Experience

ML Engineer, Kloud9.nyc (Apr 2023 - Jul 2024)
- Led Feature Engineering: Completed feature engineering and validation phases for 50+ features of a machine learning model for ghost inventory prediction at a major variety store chain, achieving 92% accuracy, contributing to 10% reduction in stock discrepancies.
- Data Analysis & Model Expansion: Analyzed ~10 datasets with multimillion data points using PySpark and Pandas to identify key statistical patterns, trends, and anomalies. Enhanced the existing ML model’s generalization, resulting in a 15% improvement in prediction stability across diverse store environments.
- Automation & Efficiency: Implemented a data analysis and insights pipeline to optimize the insights workflow, decreasing processing time from 3 hours to 20 minutes per batch.
- Cross-Functional Collaboration: Collaborated with data engineers and business stakeholders to align model objectives with business goals, streamlining the analysis pipeline and reducing deployment time by ~80%.
- Training & Mentorship: Trained and mentored 5+ interns in feature engineering, PySpark, and ML best practices, leading to 3 successful full-time conversions.
Software Engineer, WSP Global (Oct 2020 - Apr 2023)
- Developed CLI tool using PowerShell and ProjectWise to automate creation and management of employee records. Utilized by Digital Operations team for induction of ~60K new employees, saving ~2.5 human-months per year.
- Automated pipeline to detect project archival status using PowerShell, PowerAutomate and SharePoint. Developed a monthly workflow to notify project-owners and automatically archive projects. Led to 71% reduction in Digital Operations team’s maintenance overhead.
- Designed and implemented a Python application to maintain integrity in employee and project databases. Employed Pandas DataFrames to detect duplicates and inconsistencies across ~80,000 employee records per month.
Visa Official, Consulate General of France in Bangalore (Mar 2020 - Sep 2020)
- Authored 5 analysis reports on Covid-19 (w.r.t economy, education, transport), distributed to the French community of 3 states (1000+ persons); Also processed ~50 visas per day.
Management Intern, Campus France India (May 2018 - Oct 2018)
- Counselled and interviewed ~100 students for higher education. Organized 6 education fairs at 10 institutions while also training interns.

Skills

Programming Languages
- Python, R, C++, C, PowerShell, SQL, Java
Natural Language Processing (NLP)
- Hugging Face Transformers, RoBERTa, GPT APIs, Sentence Transformers, BERT-family models, Fine-tuning & Prompt Engineering, Tokenization & Embeddings, Semantic Similarity, NLI, Disagreement-aware Modeling, Evaluation & Calibration, Bias/Fairness Analysis
Machine Learning (ML) & Data Science
- PyTorch, TensorFlow, scikit-learn, MLflow, Weights & Biases Regression, Classification, Forecasting, Data Pipeline Optimization, Probability & Statistics, Linear Algebra
Systems & Tools
- Google Cloud Platform (GCP), Amazon Web Services (AWS), Docker, Flask, Git

Publications

Mandira Sawkar, Samay U. Shetty, Deepak Pandita, Tharindu Cyril Weerasooriya, Christopher M. Homan. “LPI-RIT at LeWiDi-2025: Improving Distributional Predictions via Metadata and Loss Reweighting with DisCo”. NLPerspectives workshop at Empirical Methods in Natural Language Processing (EMNLP) 2025. [Link]

Talks and Presentations

AI Alignment in Criminal Justice, RIT Graduate Showcase (Apr 2025)
- Delivered an oral presentation on AI in Criminal Justice at the RIT Graduate Showcase and awarded 1st place out of ~60 presentations; proposed a novel framework combining optimization and game theory to reduce bias in jury selection and deliberation using demographic weighting and behavioral modeling, with real-world applicability through simulations and tooling.
LLM-as-a-SupremeCourt-Judge, RIT Graduate Showcase (Apr 2025)
- Presented a poster on the LLM-as-a-SupremeCourt-Judge project at the RIT Graduate Showcase, demonstrating an AI-powered judicial simulation framework using large language models to analyze 200 U.S. Supreme Court cases from the JUSTICE dataset.

Training and Outreach

Trained and mentored 5+ interns in feature engineering, PySpark, and ML best practices, leading to 3 successful full-time conversions.
Interviewed candidates for French translation positions at WSP, leading to 4 hires. Also conducted ~20 student visa interviews for the French embassy.