Close

Ishan Saksena

Data Scientist

Download Resume

About Me

I’m Ishan Saksena, a data scientist-in-the-making fueled by curiosity, creativity, and a deep commitment to social good. For me, data isn’t just a collection of numbers; it’s a canvas for uncovering patterns, telling stories, and driving meaningful change. I thrive at the intersection of logic and imagination, where problems become opportunities for innovation. Outside of my love for data, I’m an advocate for animal welfare and a passionate storyteller. Whether I’m crafting compelling narratives or working on machine learning solutions, I believe in the power of ideas—big and small—to connect, inspire, and make a lasting impact.

Experience

University of Michigan

Graduate Student Instructor

Led a section of 25 students for SI330 Data Manipulation under Dr. Sabina Tomkins. Created assignments, midterms, and exams, conducted office hours, and led a team of Instructional Aides to support student success.

Data Science for Social Good, University of Washington

Data Science Research Intern

Analyzed 100+ million public transit records in the Puget Sound Area using PostgreSQL, Python, and R to extract insights on rider behavior and transit equity. Developed a data-driven scoring mechanism for King County Metro to prioritize bus shelter placement in underserved areas, integrating transfer and reduced fare card data. Created geospatial maps with PostGIS detailing transfer hotspots for Transit Agencies to improve service planning. Presented findings to King County, opening future research opportunities.

Augie Studio

Data Science/Analytics Intern

Spearheaded prompt engineering efforts using Amazon Bedrock to evaluate the performance of LLMs like GPT, Mixtral, Claude, and Titan. Achieved 50% cost savings by fine-tuning lower-tier models, eliminating expensive alternatives. Developed SQL scripts to enable data-driven business decisions and built analytics dashboards for monitoring user health and targeted campaigns.

Jio Platforms Limited, Mumbai, India

Data Science Intern

Developed a real-time hybrid movie recommendation system using the Neo4j Graph database, which is being put into production. Researched methods to enhance user engagement through personalized content based on social connections and user preferences. Deployed the model to a web application using Python Flask, HTML, and CSS for personalized recommendations.

Education

University of Michigan, Ann Arbor

Expected: May 2025

Master of Science in Data Science

Outstanding Masters in Data Science Student Award 2024
GPA: 4.0

Thadomal Shahani Engineering College, University of Mumbai, India

May 2023

Bachelor of Engineering in Computer Engineering

Hall of Fame 2023
Rank 1 – Semester 5, 6
GPA: 9.50/10

Projects

LMAOCaT: Low-Rank Mamba and Gated Attention Optimization

  • Developed an advanced framework starting with LLaMA 3.2, training Mamba and Gated Linear Attention (GLA) blocks to replicate softmax attention, maintaining quality while reducing compute cost.
  • Combined linear and softmax attention layers in varying ratios and arrangements inspired by Nvidia’s research.
  • Fine-tuned models with Low-Rank Adaptation (LoRA), achieving high HellaSwag benchmark scores.

Geo-political ML Study of Public Sentiment

  • Scrutinized the disjunction between official and public sentiments using Reddit data.
  • Applied ML models (Naive Bayes, Logistic Regression, XGBoost) for text classification, achieving 75% accuracy, surpassing GPT-3.5 Turbo (41%).
  • Highlighted ML’s utility in understanding public sentiment for policy insights.

Bangalore House Price Predictor

  • Built a multi-input CNN to predict housing prices, scraping data using BeautifulSoup and curating datasets.
  • Performed EDA, descriptive analysis, and data visualization using Python libraries.
  • Presented findings at ICAISC conference.

Blind-motion Image Deblurring

  • Developed an attention-based architecture achieving 92.6% accuracy for restoring blurred images.
  • Incorporated UNet-based encoder-decoder and channel attention subnetworks.
  • Integrated OCR for license plate recognition systems.

Data Science Consultant – Minarosa (Biotech Startup)

  • Conducted in-depth analysis of interview data using natural language processing (NLP) techniques to identify customer pain points and preferences, driving product and marketing strategies.
  • Performed competitor research and created a comprehensive market and patent landscape analysis for Minarosa's innovative light therapy solution for chronic UTI treatment.
  • Designed and implemented data-driven frameworks for crowdfunding campaigns and website development, leveraging insights to optimize engagement and conversion rates.

Optical Character Recognition and Machine Translation

  • Created an end-to-end application for extracting and translating text from images using Seq2Seq LSTM models and attention mechanisms.
  • Achieved a BLEU score of 31, earning the highest grade in class for this project.

Statistical Learning and Linear Regression using R

  • Implemented statistical methods in R, conducting hypothesis testing and using VIF diagnostics for model validity.
  • Applied GLS for autocorrelation and enhanced specifications with polynomial regression.
  • Analyzed data using Box-Cox transformations and studied lasso and ridge differences.

Panacea Hospital Data Volunteer

  • Collected and cleaned five years of encrypted patient data for analysis.
  • Built interactive visualizations to uncover disease patterns and seasonal trends.
  • Aided in targeted ad campaigns to increase patient visits.

Skills

Get in Touch