Electronics & Communication Engineering student at LD College of Engineering, Ahmedabad. I build end-to-end ML systems — from raw data to deployed APIs — focused on solving real problems for real people, not just running notebooks.

• Machine Learning — XGBoost, feature engineering, Optuna hyperparameter tuning, regression & classification pipelines
• Deep Learning — CNNs, RNNs, LSTMs, transfer learning (EfficientNetB0), Grad-CAM explainability
• MLOps — Flask REST APIs, Docker, Render deployment, PostgreSQL integration
• Data Analysis — EDA, hypothesis testing, Power BI dashboards

Transfer learning · EfficientNetB0 · Flask · Docker · Render

Detects disease in 15 plant-disease classes from leaf images with 93% validation accuracy. Built with a confidence threshold — returns a warning instead of a wrong prediction when the model is uncertain.

Live: plant-disease-api-l7mi.onrender.com · Repo: github.com/Kuldip-Lakhtariya/plant-disease-api

XGBoost · Optuna · Flask · Docker · PostgreSQL · Power BI · Render

End-to-end pipeline predicting whether a patient will miss their appointment. Key insight: SMS reminders appeared to increase no-shows — but the real cause was wait time (confounding variable), not SMS itself. Includes a Power BI dashboard, Flask API with HTML frontend, and full Docker deployment.

Live: medical-appointment-4gt9.onrender.com · Repo: github.com/Kuldip-Lakhtariya/Medical_appointment

Prophet · XGBoost · LSTM · PostgreSQL · yfinance

Three-model comparison for stock price forecasting with 15 technical indicators. Key lesson: high R² on raw price is misleading — the model learns the trend, not the signal. Fixed data leakage using chronological splits.

Repo: github.com/Kuldip-Lakhtariya/Reliance_stock_price_prediction

Custom CNN · BatchNorm · Dropout · Data Augmentation · TensorFlow

Built a CNN from scratch reaching 80.6% validation accuracy on CIFAR-10. Learned why transfer learning (EfficientNetB0) fails on 32×32 images — spatial collapse — and when to use a custom architecture instead.

Languages: Python, SQL
ML/DL: TensorFlow, Keras, scikit-learn, XGBoost, Optuna
Data: pandas, NumPy, PostgreSQL, SQLAlchemy
Deployment: Flask, Docker, Render, Gunicorn
Visualization: Power BI, Matplotlib, Seaborn
Tools: Jupyter, VS Code, Git, Google Colab

• NLP — text preprocessing, embeddings, sentiment analysis
• Building AI-powered web applications

• Email: kuldip2611lakhtariya@gmail.com
• LinkedIn: linkedin.com/in/kuldip-lakhtariya
• GitHub: github.com/Kuldip-Lakhtariya

I build things that work in production, not just in notebooks.