I'm the person who opens a messy CSV and gets a little excited. Null values, weird date formats, duplicated rows: that's where the story is hiding, and my job is to dig it out and tell it clearly.

• 🔍 What I do: Data Cleansing → Data Wrangling → Data Storytelling. Clean it, shape it, then make people care about it.
• 🛠️ How I do it: dashboards and analytics workflows with Python, SQL, Tableau and Microsoft Excel.
• 🚀 Currently levelling up: Python, Databases, Visual Analytics, LLMs, RAG and Agentic AI.
• 🎧 Off-screen: music, gaming, cinema. Good analysts need good recovery time.

My workflow in one line: ask a sharp question → get the data → clean it without mercy → explore → visualize → explain it so a non-technical person nods.

Because a healthy brain is a better analytical tool.

Learning in public, with receipts. Dates are in YYYY/MM/DD; 0000/00/00 means in progress. Everything verifiable lives on my Credly profile.

Courses teach concepts. Job simulations make you do the job. These two put me in the shoes of a real analyst working on real business questions.

Four projects, four different businesses, one recurring question: "So what should someone actually do about this?"

All four are vibe-coding builds. I act as the architect (business questions, data logic, what the dashboard must answer) and let AI handle the typing. Every one of them is deployed, so you can click and explore instead of cloning and praying.

📍 Code lives in: Himanshu-5680/Data-Analysis_Projects → Vibe_coding_Projects

The story: I got one CSV: 121,445 rows × 56 columns. It looked like a single table, but it wasn't. It was seven separate business tables stacked on top of each other (orders, delivery, feedback, customers, inventory, campaigns, products), and each row only made sense if you knew which columns were filled in.

So the first job was detective work. I found the split by studying which columns were populated together, un-merged the file into seven clean sub-tables, and checked the math: the row counts added up to exactly 121,445. Orders turned out to be stored as two row types (headers and items) that had to be joined back on order_id. All 5,000 matched.

What the data said:

• 📦 5,000 orders, ₹1,10,09,309 revenue, ₹2,202 average order value
• ⏱️ 69.4% on-time deliveries, so roughly 3 in 10 orders ran late. Delays track with lower ratings and worse sentiment.
• ⭐ Average rating 3.34 / 5, with sentiment split almost evenly three ways
• 📣 Marketing ROAS of 1.97x. Email leads (2.05x), App trails (1.92x).
• 🧴 Household Care and Personal Care lose 51–57% of inventory to damage, the biggest supply-chain leak in the data

The honest part: the data has quirks (order totals don't equal quantity × price, and about 31% of inventory rows show damage above stock received). I flagged them in the dashboard and never silently "fixed" them. Trust in an analysis starts there.

The dashboard: seven tabs (Overview, Orders & Revenue, Delivery, Feedback, Marketing, Inventory, Customers) with global sidebar filters, all computed live from the data with no hard-coded numbers.

🔗 Live app · Code

The story: every company loses people, and most dashboards just count how many. I wanted one that answers who is at risk, why, and what HR should do about it.

I took IBM's HR dataset (1,470 employees) and built a five-level decision surface:

1. KPIs: what is happening?
2. Trends: which direction is it moving?
3. Drivers: why? (overtime, role, department, marital status, commute distance)
4. Risks & Opportunities: what's next? A risk model, a high-risk employee list, and KMeans segments
5. Strategic Actions: what should HR do? Every action is tied to a specific chart.

Under the hood: a balanced Logistic Regression baseline, then a GridSearchCV-tuned Random Forest, judged on F1 / ROC-AUC / Precision / Recall, not vanity accuracy. A second model predicts income. Constant columns and ID fields were dropped up front.

A design choice I care about: model output is positioned as a prioritisation aid for human review, never an automated employment decision. Predicting people's outcomes deserves that caution.

Every level ends with a Key Insights callout, and there's a one-click PDF export of the live dashboard.

🔗 Live app · Code

The story: what does Netflix's catalog say about its strategy? I used the public shivamb/netflix-shows dataset and looked at it the way a content strategist would: what's being added, where's the catalog aging, and where are the gaps?

The app follows the same five-level BI framework (KPIs → Trends → Drivers → Risks & Opportunities → Actions):

• 📈 Year-over-year acquisition trajectory and monthly seasonality
• 🎭 Top genres, movie runtime distribution, and TV season retention
• 🌏 Catalog aging, single-season churn, and APAC / LATAM whitespace
• 🧭 A strategic action matrix

One detail I'm proud of: the dataset's rating field is a maturity label (TV-MA, TV-14, PG-13...), not a quality score. I refused to turn it into a fake IMDb-style rating and grouped it into audience segments instead. Bad assumptions make confident-looking, wrong charts.

The whole platform is one app.py, with data-quality checks built in and a Word report (.docx) generated on demand from the current filtered data.

🔗 Live app · Code

The story: a multi-branch supermarket chain has sales data but no insight. Managers can't quickly answer which branch wins, which category pays the bills, and when to plan for peaks. This project turns 500 transactions across four branches (Jaipur, Delhi, Mumbai, Bengaluru) into a six-tab dashboard.

This one is also my purest Vibe Coding experiment: I framed the business questions, specified the architecture (a clean split between analysis.py for data and app.py for UI), and directed IBM Bob to build it, including debugging a real TypeError and generating a 12-slide PowerPoint report for the business team.

What the data said:

• 💰 ₹2,44,411 total revenue across 500 transactions
• 🏆 Mumbai (Branch C) is the top branch at ₹72,469
• 🥤 Beverages lead categories (22.96% of sales); Cheese is the top product
• 📱 UPI is the top payment method (27.8%), and over half of revenue is fully digital
• 🎫 Members drive most of the revenue, so the loyalty programme matters
• 📅 April 2026 was the peak month; Bakery is the untapped opportunity

The output isn't just charts: it ends with prioritised business recommendations (High / Medium / Low) and a Product × Branch heatmap for deeper digging.

🔗 Live app · Code

Python is my daily driver for cleaning, exploring, testing and explaining data.

• Data Analytics Essentials (Cisco Networking Academy)
• Enterprise Data Science in Practice (IBM SkillsBuild)
• Python Essentials 1 & 2 (Cisco Networking Academy, in progress)

Fundamentals: variables, data types, strings, operators, conditionals, loops, functions, lambda, list comprehensions, data structures, file handling, exception handling, and lots of Jupyter Notebook time.

Applied work:

• Cleansing, wrangling and transformation with pandas & numpy
• Visualization and statistical plotting with matplotlib & seaborn
• Data classification and sentiment-analysis workflows

If Python is the workshop, SQL is the warehouse where the data lives.

• Data Analytics Essentials (Cisco Networking Academy)
• Databases and SQL for Data Science with Python (IBM / Coursera, in progress)

Comfortable with: relational database fundamentals, schemas and tables, SELECT, WHERE, ORDER BY, GROUP BY, HAVING, aggregation, and joining tables (INNER JOIN, LEFT JOIN, and friends).

From raw data to executive dashboard, end to end.

• Data Fundamentals & Getting Started with Data (IBM SkillsBuild)
• Deloitte Australia Data Analytics Job Simulation (Forage)
• Quantium Data Analytics Job Simulation (Forage)
• Visual Analytics with Tableau (UC Davis / Coursera, in progress)

Core strengths: Data Analysis · Data Analytics · Data Engineering fundamentals · Big Data concepts · Data Cleansing · Data Wrangling · Data Literacy · Data Visualization · Data Storytelling · interactive dashboards in Tableau & Microsoft Excel.

The future analyst works with AI, not against it.

• Generative AI Essentials: Using LLMs to Work with Data (IBM SkillsBuild)
• Make Agentic AI Work for You (IBM SkillsBuild)

Concepts: Large Language Models (LLMs) & prompt engineering for data workflows · Retrieval Augmented Generation (RAG) · AI Agents & multi-agent task automation · Sentiment Analysis & IBM Watson Knowledge Studio (WKS).

IDE = Integrated Development Environment

Tool	Why it's in my toolkit
	Exploratory data analysis: code, output and notes in one place.
	Scripting, project organization and Git integration.
	Cloud Python and easy notebook sharing on GitHub.

CLI = Command Line Interface

1. Command Prompt
2. Windows PowerShell
3. Git

My personal cheat sheets, written so future-me (and you) can skip the googling.

• Basic writing and formatting syntax
• emoji-cheat-sheet
• Welcome! Badges 4 README.md Profile
• shields.io
• Simple Icons