~/yash-saini |

I build analytics systemsthat turn messy data into decisions.

Data Analyst | BI Engineer | Building pipelines, analytics systems, and dashboards that drive decisions

I'm Yash Saini. I don't collect tools. I ship outcomes. At CUNY, I built pipelines across 26 institutions that turned weeks of reporting into minutes. At i3 Infosoft, I segmented 1M+ CRM records and the marketing team rewrote their entire quarterly strategy based on what the data showed. I'm finishing my MS in Business Analytics at Baruch (Zicklin) and looking for a team where the data actually changes something.

faster reporting

institutions served

accuracy gain

KPI dashboards

See what I'd build for your company Resume PDF

Yash Saini, Data Analyst and BI Engineer based in New York City

// live_demonstration

What I'd Build At Your Company

Type any company name. I'll detect your industry, map data challenges, design an architecture, and draft a 90-day implementation plan. Every output is different. Try it twice.

company_intelligence.py

// the_point

Everything you just experienced was designed and built by me. The industry detection, the architecture generation, the tailored proposals. Each company gets a different output because the logic adapts, not because I memorized answers.

The platform is the proof. You just experienced my skills.

SQL & BigQueryPythonETL PipelinesData ModelingBI DashboardsA/B TestingForecastingWeb Development

// featured_case_study

MTA Turnstile Data Engineering

The full story of how I turned millions of messy transit records into a queryable, automated analytics system on Google Cloud.

From Raw MTA Files to Real-Time Transit Intelligence

END-TO-END PROJECT

The Problem

NYC's subway system has 472 stations and thousands of turnstiles, each recording cumulative entry/exit counts every 4 hours. The MTA publishes this data weekly as flat text files. But the raw data is riddled with problems: counter resets that produce negative values, duplicate timestamps, inconsistent station naming, and no documentation on edge cases. Analysts who want to answer basic questions ("Which stations are busiest on Friday evenings?") end up spending hours in Excel cleaning the same mess every time.

Data Scale & Challenges

Millions of rows per quarter. Each row represents one turnstile's reading at one timestamp. The cumulative counters mean you need to calculate deltas between readings, but counters reset unpredictably, creating massive negative spikes. Station names vary across files ("42 ST-TIMES SQ" vs "TIMES SQ-42 ST"). Some turnstiles report at non-standard intervals. All of this had to be handled before any analysis could begin.

Architecture & Pipeline

I designed a 5-layer cloud pipeline on Google Cloud Platform:

📥

MTA Raw Files

🔄

Kafka Stream

☁️

BigQuery

⚙️

dbt Models

📊

Analytics

Kafka handles streaming ingestion. BigQuery stores the warehouse. dbt manages transformations as version-controlled SQL models. Prefect orchestrates the whole thing on a schedule, so new weekly data flows through without manual intervention.

Key Transformation Logic

The hardest part was computing reliable ridership from cumulative counters. Here's the core logic:

-- Calculate per-turnstile ridership deltas -- Filter out counter resets and anomalies WITH deltas AS ( SELECT station, unit, scp, datetime, entries - LAG(entries) OVER( PARTITION BY station, unit, scp ORDER BY datetime ) AS entry_delta FROM raw_turnstile ) SELECT station, datetime, SUM(entry_delta) AS ridership FROM deltas WHERE entry_delta BETWEEN 0 AND 10000 GROUP BY station, datetime

The BETWEEN 0 AND 10000 filter catches counter resets (negative values) and anomalous spikes (readings above 10K per 4-hour window are physically impossible for a single turnstile).

Impact & Insights

What used to take hours of manual Excel work now runs automatically in under 5 seconds. The pipeline processes millions of records end-to-end without intervention. Ridership patterns became instantly queryable: peak hours, seasonal trends, station comparisons, all accessible through simple SQL against a clean warehouse. The project demonstrates the full data engineering lifecycle: ingestion, streaming, warehousing, transformation, orchestration, and analytics.

// selected_work

Projects That Prove It

Click any project to see findings, interactive charts, and results explained for anyone to follow.

📊

Customer Churn Prediction

Built a model that flags customers about to cancel with 84% accuracy. Connected it to a live dashboard so retention teams can act the same day, not the same quarter.

PythonScikit-learnSQLPower BI

Click for findings & charts

// in plain english

Imagine running a subscription business where 1,000 customers are about to leave but you don't know which ones. This model identifies them before they cancel, so retention teams can reach out with the right offer at the right moment.

// key findings

🎯

84% accuracy identifying at-risk customers before they cancel

📉

Month-to-month contracts churn at 3.5x the rate of annual subscribers

⏱️

62% of all churn happens within the first 6 months

💰

Estimated $140K annual savings if retention team acts on predictions

// feature importance

View code on GitHub

🏙️

NYC Housing Equity Analysis

Used data from 4 city agencies to test whether low-income neighborhoods get slower responses to housing complaints. The answer: yes, measurably and consistently.

RQuartoRandom ForestSpatial Joins

Click for findings & charts

// in plain english

When you call 311 about broken heat or mold in NYC, how fast the city responds should not depend on your zip code's income level. This analysis found it does. Lower-income neighborhoods wait significantly longer, and the pattern holds even after controlling for complaint volume.

// key findings

📊

R-squared = 0.54: income alone explains over half the variation in resolution time

⏱️

Low-income tracts wait 40% longer on average for complaint resolution

🗺️

4 public APIs combined with spatial joins across 100K+ complaints

🔬

Pattern is statistically significant even after adjusting for complaint type and volume

// income vs resolution time

View code on GitHub

🛒

Retail Analytics Pipeline

Took chaotic Excel sales data and turned it into a clean, queryable database with interactive dashboards. The full journey from raw mess to business insight.

PythonSQLSQLitePower BI

Click for findings & charts

// in plain english

Most small businesses have sales data trapped in messy spreadsheets nobody can read. I built a system that cleans, organizes, and visualizes this data, turning chaotic files into clear dashboards where you can instantly see what sells, what doesn't, and where to focus.

// key findings

🧹

Raw Excel to structured database with automated cleaning for nulls, duplicates, formatting

🏗️

Normalized relational model with proper fact and dimension tables in SQLite

📈

Interactive Power BI dashboard tracking sales trends, top products, and customer segments

// monthly revenue trend

View code on GitHub

// methodology

How I Approach Data Problems

Every project follows the same discipline. The tools change, the thinking doesn't.

🎯

Understand the Decision

What action will this analysis inform? Who needs it, and when? I start with the business question, not the data.

🔍

Investigate the Sources

Map every data source end-to-end. Where does it break? What's missing? What's inconsistent? This step saves weeks later.

🔧

Build Reproducible Pipelines

No one-off scripts. Automated, validated, version-controlled pipelines that run without me babysitting them.

📊

Analyze & Test Hypotheses

Segmentation, modeling, experimentation. Let the data answer the question, not confirm a hunch.

🚀

Deliver Decision-Ready Insights

Dashboards that stakeholders can use without analyst support. If the output needs a tour guide, it's not done.

// where_ive_built

Work Experience

Data Specialist Intern

Research Foundation of CUNY, New York

Aug 2025 - Present

55%faster reporting

37%accuracy gain

26institutions

7+sources unified

Built the SQL and BigQuery pipelines that consolidated fragmented reporting across all 26 CUNY institutions. Designed validation frameworks that caught data integrity issues two cycles before a major deadline. Automated ETL that gave operations teams self-serve dashboards for the first time.

Founder & Growth Analyst

Gharstuff, E-commerce

Feb 2020 - Present

65%reach growth

55%signups up

45%conversion lift

Built a category-based e-commerce platform from zero and ran it as a one-person data operation. Designed customer acquisition cohorts to track which referral channels produced repeat buyers vs. one-time visitors. A/B tested every paid and organic campaign, iterating weekly on messaging based on behavioral data. Built referral funnel analytics that identified drop-off points, then restructured the flow to increase sign-ups 55%. Improved conversion rates 45% through audience segmentation. Every decision was tied to a number because there was no budget for guessing.

Data Analyst

i3 Infosoft Pvt. Ltd., Noida, India

Sep 2022 - Jul 2024

1M+CRM records

18%engagement lift

20%forecast accuracy

15+KPI dashboards

Ran A/B experiments and built Python segmentation models across the entire CRM. The 18% engagement lift became the basis for the next quarter's channel strategy. Replaced manual reporting with automated SQL pipelines. Built time-series forecasting that improved inventory planning by 20%.

// how_i_work

Skills, With Receipts

No percentages. Here's what I've actually shipped with each tool.

SQL & BigQuery

Built pipelines serving 26 institutions. Migrated 7+ fragmented sources into normalized structures. Cut query runtimes 25%.

Python

Segmented 1M+ CRM records. Built time-series forecasting. Automated ETL replacing hours of manual work.

Power BI & Tableau

Designed dashboards tracking 15+ KPIs that moved ops from weekly reports to same-day visibility.

GCP & Cloud

Deployed unified reporting layer on Google Cloud. Built BigQuery warehouse for MTA analytics.

R Programming

Spatial analysis across 100K+ NYC records from 4 public APIs. Random Forest with R-squared 0.54.

A/B Testing

Designed experiments producing 18% engagement lift. Ran campaign optimization across organic and paid.

ETL & dbt

Orchestrated pipelines with Prefect and dbt. Kafka streaming for real-time ingestion.

Data Governance

Built validation frameworks catching integrity issues before deadlines. Standardized metrics across 26 institutions.

I build analytics systemsthat turn messy data into decisions.

I'd rather show you than tell you.