AI/ML Undergraduate · Data Scientist
Himanshu Gupta builds with data.
I build end-to-end analytics and machine learning systems, from messy datasets to models people can actually trust and dashboards people actually use.
About
What I work on, and why data ends up at the center of it.
I'm pursuing my B.Tech in AI & Machine Learning, but my journey actually started with web development — I was mostly curious about building websites and what happens behind the scenes.
That curiosity drifted toward data, and AI was booming at the time too. It eventually led me to Python, SQL, and data analysis, and from there into statistics and machine learning.
Since then I've learned mostly by building — working on projects around data analytics, customer behaviour, and machine learning. I've since gone through supervised & unsupervised learning, deep learning fundamentals, and RNNs, and I'm now going deep into NLP as my primary focus — aiming for real theoretical understanding, not just applied tool-use.
- Based inNew Delhi, India
- FocusMachine learning & NLP systems
- CurrentlyB.Tech AI & ML, VIPS · 2023–2027
Skills
What I build with.
The languages, tools and concepts I actually work with day to day.
Languages & Data
ML & Libraries
NLP & Deep Learning
Projects
Selected work.
Five projects that show the full loop — from raw data to a decision someone downstream can use.
Gutenberg Next-Word Predictor
A word-level next-word prediction language model built with PyTorch and trained on cleaned Project Gutenberg text.
- Built a full pipeline from text cleaning and tokenization to an Embedding + stacked LSTM language model.
- Used fixed-length context windows, cross-entropy loss, and early stopping based on validation loss.
- Reached 15.86% top-1 and 33.99% top-5 validation accuracy, with GPU acceleration when available.
- Added top-k sampling for less repetitive generation, and saved model/tokenizer checkpoints for reuse.
Email/SMS Spam Classifier
An end-to-end NLP pipeline that classifies messages as spam or ham, from raw text preprocessing to a deployed web app.
- Preprocessed text with tokenization, stopword removal, and stemming; compared CountVectorizer and TF-IDF features.
- Evaluated Decision Trees, Random Forest, and three Naive Bayes variants on Accuracy, Precision, Recall & F1-score.
- Best model — TF-IDF + Bernoulli Naive Bayes — reached 98.84% accuracy and 95.24% F1-score.
- Deployed as an interactive Streamlit web app for real-time message classification.
Customer Satisfaction (CSAT) Prediction
An end-to-end pipeline that flags dissatisfied customers before they escalate, using structured and textual features.
- Compared Logistic Regression, Random Forest, Decision Tree & XGBoost with SMOTE + GridSearchCV tuning.
- Best model reached 0.804 F1-score and 79.2% accuracy on XGBoost.
- Used SHAP to explain predictions and surface the key drivers of satisfaction.
- Shipped as a real-time Streamlit app for scenario testing.
Customer Behaviour & Purchase Decision Analysis
A multi-page Power BI dashboard uncovering what actually drives revenue and returns across customer segments.
- Found high-income adult customers drive 50%+ of total revenue.
- Non-loyal customers spent ₹27 more per purchase — a loyalty-program red flag.
- Ad engagement outperformed discount campaigns on ROI, per DAX measures.
- Returns showed no correlation with satisfaction — pointing to product/ops causes.
The Three-Point Revolution: NBA Evolution (2001–2025)
A multi-page Power BI dashboard analyzing two decades of league-wide data to uncover the structural shift toward perimeter-focused offense.
- Cleaned and modeled 2001–2025 NBA team & player data with Python and SQL, including window functions for season-over-season trends.
- Found three-point attempt rate accelerated sharply after 2014, while shooting efficiency stayed stable.
- Classified team strategies (Balanced, Paint-Focused, Three-Point Heavy) and linked strategy type to win %.
- Built a 5-page interactive dashboard covering league trends, player role specialization, and strategy evolution.
Resume
Experience & education.
B.Tech — Artificial Intelligence & Machine Learning
Vivekananda Institute of Professional Studies, New Delhi
Contact
Let's talk data.
Open to Data Analyst and Data Science internship opportunities — reach out any time.
Send me an email
himanshu.13122003@gmail.com
Click this box to open a pre-filled message in Gmail.
Opens in a new tab