Data Analyst · Data Science · Machine Learning
I turn messy, inconsistent data into dashboards and recommendations people actually act on. Python, SQL, Power BI, and applied LLM / fine-tuning work.
At Carrier Compiler, I built Power BI dashboards tracking enrollment, revenue, and performance across a multi-course ed-tech catalog — using SQL to extract and query the data and Excel to clean it before it reached a chart.
On my own projects I go end-to-end: a churn dashboard on 7,043 telecom customers that isolated a ~65% churn high-risk segment and produced six prioritized retention recommendations; a two-page Superstore sales dashboard with a 15-day forecast; and an exploratory analysis of 5,000 movies testing whether critical acclaim predicts revenue (it doesn't — ~0.03 correlation; budget is the real driver).
I also spent two years teaching Maths and CBSE Computer Science to students aged 11–18, which is where I actually learned the skill that matters most in this work: explaining what a number means to someone who doesn't want the methodology, just the answer.
Outside core analytics, I build with LLMs and fine-tune small models (LoRA/PEFT) — not to chase the hype, but because understanding how the models work makes me faster at knowing when not to reach for one.
End-to-end Power BI dashboard on 7,043 IBM Telco customers. Engineered tenure/charge-band features, authored DAX for churn & retention rate. Found month-to-month contracts churn at 43% vs. 3% for two-year contracts; isolated a ~1,100-customer, ~65%-churn high-risk segment and delivered 6 prioritized retention recommendations.
Two-page Power BI dashboard on 5,901 Superstore transactions (2019–2020) with a 15-day sales forecast and region-level profit analysis. Solved real data-prep issues — locale date formats and #N/A handling — for a clean, refresh-ready model.
End-to-end exploratory analysis on the TMDB 5000 Movies dataset, parsing nested JSON columns (genres, cast, crew). Tested whether critical acclaim drives box-office success — found a weak ~0.03 correlation, with budget the strongest predictor — and shipped the finding as a structured report with 5+ visualizations.
A SQL-based analysis of global layoff trends across companies, industries, stages, and countries — a complete workflow from raw data to cleaned, query-ready tables and findings.
Tableau-based BI project analyzing Airbnb listing performance, pricing patterns, and revenue trends, including the workbook and source dataset.
Professional Excel-based BI project analyzing bike purchase behavior across customer demographics, income levels, education, region, and product ownership patterns.
Collects and prepares Formula 1 race-lap data via a FastF1 pipeline, then evaluates an XGBoost model for lap-time prediction.
Lightweight educational fine-tuning pipeline for Qwen3 0.6B. Reads source documents from PDF, TXT, DOCX, and EPUB, cleans and normalizes the text, and builds a supervised chat dataset.
A four-stage multi-agent research pipeline: a search agent, a reader/scraper agent, a writer chain, and a critic chain work together to turn a topic into a structured report.
Image-based Automatic Number Plate Recognition (ANPR) for vehicle detection, license plate localization, and OCR, with a Streamlit dashboard and a YOLOv8 training utility.
Employee attrition analysis on a structured HR dataset, visualizing key workforce trends through a Power BI dashboard.
Sales and business performance dashboard for an e-commerce dataset, combining order-level transaction details with customer and geographic metadata.