| Heart Disease (Cleveland) | |
|---|---|
| 📁 Samples | 303 |
| 🏷️ Features | 13 (age, cholesterol, ECG…) |
| 🎯 Classes | 2 (presence/absence) |
| 📅 Introduced | 1988 · UCI · Detrano et al. |
| 💡 Fun fact | Despite only 303 samples, it's been cited 3,000+ times — proof that a tiny but well-curated dataset can outlast massive ones |
No disease ███████████░░░░░░░░░ 54%
Disease █████████░░░░░░░░░░░ 46%
⚙️ Auto-updates daily to spotlight a new dataset. Check back tomorrow!
MS Data Analytics student at UIS and ex-Data Analyst at Tredence Analytics (2+ yrs, Fortune 500 client). I build end-to-end ML pipelines, predictive models, and data visualizations — from churn models driving $4M/month revenue to fine-tuning transformers with QLoRA on a single GPU.
- 💼 Ex-Data Analyst @ Tredence Analytics · Bangalore, Karnataka (India)
- 🥽 Graduate Research Assistant (VR Data Visualization Research | Meta Quest 3 + Unity)
- 🎓 Master of Science in Data Analytics · University of Illinois at Springfield (2024–2025)
- 🎓 Bachelor of Technology in Computer Science & Engineering · National Institute of Technology, Hamirpur (2018–2022)
- 📍 Based out of Springfield, IL (USA)
ML & Deep Learning
Languages & Scripting
Data & Visualization
Engineering & Cloud
Design & XR
| Project | Description | Stack | Result |
|---|---|---|---|
| 🌍 VasuDev | Global news intelligence platform — real-time pulse map of every country with multi-perspective AI briefings, freeform chat, and historical Time Machine | React · Leaflet · FastAPI · Supabase · Groq | PWA · 195 countries |
| 📋 OpenReturn | Free AI-powered conversational assistant for US federal tax guidance. Uses RAG over IRS documents (2023–2025) + Llama 3.3 70B via Groq to provide plain-English line-by-line help and generate a personalized PDF roadmap. | Streamlit · LangGraph · LangChain · ChromaDB · Groq (Llama 3.3 70B) · ReportLab | Live on Hugging Face Spaces |
| 🤙 Formal → Gen Z Translator | Fine-tuned FLAN-T5 with QLoRA on 28k pairs; injected 210 emoji/slang tokens | HuggingFace · QLoRA · W&B | BLEU 42→58 · ROUGE-L 0.54→0.76 |
| 🧮 GSM8K Math Reasoning | Fine-tuned FLAN-T5 with 4-bit QLoRA on 8.5k math word problems | HuggingFace · QLoRA · BitsAndBytes | Exact match eval on GSM8K |
| 📈 Time Series Forecasting | Oil temp prediction on ETTh1 — LSTM, 1D CNN, and N-BEATS comparison | TensorFlow · N-BEATS | MAE 0.76°C vs 1.99°C naive baseline (62% ↑) |
| 🖼️ Image Segmentation U-Net | Custom U-Net + VGG16 encoder for multi-class semantic segmentation | TensorFlow · U-Net · VGG16 | MeanIoU across 8 classes |
| ✂️ RPS Classifier (CNN) | 5 CNN architectures + DenseNet121 transfer learning for RPS classification | TensorFlow · DenseNet121 | ~99% val accuracy |
| Two-layer → multi-layer ANN built with TensorFlow primitives; flight price regression | TensorFlow · GradientTape | RMSE vs Linear Regression baseline | |
| 🔢 MNIST ANN from Scratch | Multi-class ANN using raw TF ops — GradientTape, matmul, He init, softmax | TensorFlow · GradientTape | ~91% val accuracy |
| ⚡ MNIST ANN — Improved | Same ANN upgraded with Adam, mini-batch GD, BatchNorm, dropout, L2, plateau LR | TensorFlow · Adam · BatchNorm | ~96% val accuracy |
| 🧪 MNIST with Keras | Sequential API progression: baseline → BatchNorm → dropout → Hyperband tuning | Keras · Keras Tuner · Hyperband | 98.3% val accuracy |
| ⚖️ Recidivism Prediction | Lasso, Ridge, RF, GBM, SVM, NN + Shapley values & fairness metrics | R · caret · glmnet · keras3 | RF best at 79.6% · fairness analysis |
| 🐘 Guess the Animal Game | Interactive ML game — model IDs your animal via yes/no questions (AWA2, R + Python) | Scikit-learn · Decision Tree · R | Clean tree traversal game loop |
| 📰 Fake News Detection | Naive Bayes vs ANN comparison for fake news classification | R · Keras · NLP | Naive Bayes 91.1% |
| 📣 Marketing Campaign KNN | KNN & Naive Bayes for campaign outcome prediction | R · Scikit-learn | KNN 75% · NB 72% |
"You are one observation in an infinite dataset. Yet the universe only knows itself through observations like you, which makes you, somehow, all of it. ॐ "