sumit_prajapat=# SELECT * FROM about WHERE honest = true;I take data the whole way: ingestion → Azure pipelines → SQL star schemas → Power BI dashboards → ML models — and I lead every dashboard with one quantified headline, not a wall of charts.
2+ years inside BFSI selling and servicing insurance products before analyzing them — now a freelance Data Analyst with Finonus Capital. B.Com graduate, Microsoft (PL-300 · AZ-900 · DP-900) and AWS certified.
-- The kind of analyst who measures e-commerce retention on customer_unique_id — the real person —-- and flags a dataset cut-off artifact before anyone misreads it as a downturn.-- and says "the data can't prove this campaign worked" when a placebo test fails.
Domains Retail & FMCG Finance & Insurance Healthcare E-Commerce · Seeking Data Analyst BI Developer Power BI Developer
sumit_prajapat=# EXPLAIN ANALYZE how_i_ship;
sumit_prajapat=# SELECT project, headline_metric
sumit_prajapat-# FROM featured_work
sumit_prajapat-# ORDER BY impact DESC;# |
project |
headline_metric |
|---|---|---|
01 |
Grocery Retail Analytics NEW |
1.46M transactions → 50.6% of coupons go to habitual buyers |
02 |
Healthcare Insurance Claims Analytics | 558,211 claims → 226× fraud concentration |
03 |
Mutual Fund Analytics Platform | 14,384 schemes · 51 AMCs · Prophet forecasts |
04 |
Olist E-Commerce Analytics | 112,650 records → ~97% one-time buyers exposed |
05 |
Customer Segmentation Engine | RFM + 3 models · Silhouette 0.52 · live app |
06 |
Queryscratch | 500 SQL questions · in-browser SQL sandbox |
(6 rows)
sumit_prajapat=# \x on -- expanded display: full detail per project
|
| 1.46M transactions |
50.6% habitual coupons |
225% display+mailer lift |
58/58 DAX = SQL |
24-check data-quality gate → DuckDB star schema → 13 SQL analyses → difference-in-differences with placebo tests → 5-page Power BI report (44 DAX measures) + Streamlit app with a cite-or-refuse “ask the data” assistant.
- Top 10% of households drive 34.2% of sales; 318 lapsing regulars hold 13.4%
- Placebo checks failed → campaign ROI unmeasurable; recommended a 10% random holdout
- SQL-vs-DAX reconciliation caught 3 DAX bugs before they reached the report
Python DuckDB SQL statsmodels Power BI DAX Streamlit GitHub Actions
558,211 Medicare claims · 5,410 providers · where does the fraud concentrate?
| 558K claims |
226× fraud concentration |
90.1% fraud recall |
$295.7M exposure flagged |
Azure Blob → Data Factory → PostgreSQL 9-table warehouse (2.43M rows) → scikit-learn fraud model (AUC 0.9573) → 4-page Power BI report with provider drill-through.
- Top billing quartile: 34.02% fraud rate vs 0.15% in the bottom quartile
- 96% of high-velocity providers (>10 claims/day) are fraud-flagged
- Fraud providers file 9+ diagnosis codes 1.66× as often — diagnosis stuffing
Azure ADF PostgreSQL Python scikit-learn Power BI DAX
14,384 schemes · 51 AMCs · one risk framework for every fund
| 14,384 fund schemes |
51 AMCs |
5 + 4 dim + fact tables |
25+ DAX measures |
Yahoo Finance + AMFI → Azure Blob → Data Factory → Azure SQL star schema → risk suite in Python → Prophet NAV forecasts → 4-page Power BI report and a live Streamlit app.
- Full risk suite: Sharpe, Sortino, Treynor, alpha/beta and max drawdown
- 30/60/90-day Prophet NAV forecasts with confidence bands
- Azure Data Factory pipeline running end to end into a star schema
Azure ADF Azure SQL Python Prophet Power BI Streamlit
112,650 sales records · 96,096 real customers · why don't buyers come back?
| 112,650 sales records |
~97% one-time buyers |
91.88% on-time delivery |
40+ DAX measures |
8 raw tables → Azure Blob → Data Factory → Azure PostgreSQL star schema (4 dims + 1 fact) → 3-page Power BI report with decomposition tree, key influencers and what-if price simulation.
- Retention measured on customer_unique_id — only ~3% of real customers ever reorder
- São Paulo drives 38.3% of revenue; health_beauty leads at R$1.4M
- Remote AP state waits 28 days vs a 21-day average delivery
Azure ADF PostgreSQL SQL Power BI DAX Power Query
~541K retail transactions · RFM + 3 clustering models · who is worth keeping?
| 541K transactions |
0.52 silhouette score |
~40% revenue from top 15% |
£127K recoverable revenue |
UCI Online Retail data → cleaning → RFM scoring → K-Means vs Hierarchical vs DBSCAN benchmarked head-to-head → PCA, CLV and cohort analysis → interactive Streamlit app.
- Champions are 8× the lifetime value of lost customers
- Cohorts lose 70%+ of customers by month 3 — the window to act
- K-Means (K=4) beat Hierarchical and DBSCAN on compact, explainable clusters
Python Pandas scikit-learn PCA Streamlit
500 SQL challenges · runs in the browser · free for students
| 500 SQL challenges |
0 installs needed |
WASM in-browser engine |
Free for students |
A free SQL learning platform: hand-crafted challenges from a first SELECT to recursive CTEs, with real queries run client-side via sql.js / WebAssembly and instant exact-output feedback.
- No server round-trip — every query runs inside the browser
- Learn, playground, mock-interview and practice tracks, beginner → expert
- Streaks and per-user progress on Supabase auth
SQL sql.js WebAssembly Supabase JavaScript
sumit_prajapat=# SELECT category, tools FROM toolkit GROUP BY category;category |
tools |
|---|---|
business_intel |
Power BI DAX Power Query Excel |
data_engineering |
SQL PostgreSQL DuckDB Azure Data Factory Star Schema ETL |
python_ml |
Python Pandas NumPy scikit-learn statsmodels Prophet Streamlit |
cloud_and_ops |
Microsoft Azure AWS Git GitHub Actions Jupyter |
sumit_prajapat=# SELECT * FROM certifications WHERE verified = true;credential |
issuer |
status |
|---|---|---|
| ★ PL-300 · Power BI Data Analyst Associate | Microsoft | → verify |
| AZ-900 · Azure Fundamentals | Microsoft | → verify |
| DP-900 · Azure Data Fundamentals | Microsoft | → verify |
| Cloud Practitioner | AWS | → verify |
| Investment Foundations | CFA Institute | → verify |
| SQL Certified | HackerRank | → verify |
