Developed and deployed an AI pipeline integrating computer vision and natural language processing for a Visual Question Answering (VQA) task.
The system used transformer-based models and reinforcement learning (PPO) to improve reasoning and response accuracy.
Optimized model latency to 392ms/query and deployed the solution as a reproducible Hugging Face demo with an interactive dashboard interface.

🧩 Tech Stack:
Python · PyTorch · Hugging Face · Flask · MAG · Reinforcement Learning(PPO)

🎯 Outcomes:

Achieved 81.4% (Strict) and 75.6% (VQA2) accuracy on benchmark datasets
Released production-ready open-source demo

ChurnChampion is a predictive analytics project designed to forecast customer churn and help businesses improve retention strategies. The project includes data preprocessing, feature engineering, model training, and interactive visualization through a Dash web app.
Machine learning models such as Logistic Regression, Random Forest, and XGBoost were used to predict churn probability, while the dashboard allows users to visualize model performance, feature importance, and customer behavior trends.

Key Features:
End-to-end machine learning pipeline for churn prediction
Interactive Dash web app for exploring churn patterns and model metrics
Visualizations: Confusion Matrix, ROC Curve, Feature Importance, and Churn Distribution
Reusable project structure for training, evaluation, and reporting

Tech Stack:
Python · Scikit-learn · Dash · Plotly · Pandas · NumPy · Seaborn · Matplotli

Developed an OCR pipeline using CNN + CRNN architectures, achieving 92% accuracy on a 200K-image dataset. Optimized training using augmentation, transfer learning, and early stopping, reducing model training time by 25%. Deployed for 10K+ active users to automate document processing.

Tech Stack: Python · TensorFlow · OpenCV · CNN/CRNN

Designed a collaborative + content-based recommendation engine for a learning platform, boosting user engagement by 20%. Automated ETL pipelines and reporting workflows, reducing manual data effort by 30% and improving personalization accuracy.
Tech Stack: Python · Pandas · NumPy · Matplotlib

Roles: Data Scientist · ML Developer

Created time-series forecasting models (ARIMA, LSTM) to predict stock trends using 5 years of historical data. Built an interactive visualization dashboard for investors to explore predictions, trends, and confidence intervals.

Tech Stack: Python · Power BI · Matplotlib · Pandas

Developed an NLP pipeline using LSTMs and Transformer models on 50K+ customer reviews to classify sentiment and extract key insights. Built dashboards to visualize sentiment trends and feedback categories for business intelligence. Achieved 85%+ classification accuracy.

Tech Stack: Python · PyTorch · Hugging Face · Matplotlib

Applied K-Means clustering and PCA on a 30K+ customer dataset to identify marketing segments and improve campaign targeting. Built a Power BI dashboard to visualize clusters and KPIs across demographics, purchasing patterns, and engagement levels.

Tech Stack: Power BI · Python (Pandas, Scikit-learn) · SQL

Developed supervised ML models (Logistic Regression, Random Forest) and anomaly detection pipelines for fraud detection using 200K+ transactional records. Performed feature engineering and evaluated performance with confusion matrix, precision, and recall metrics. Visualized fraud patterns and risk scores for business users.

Tech Stack: Python · Scikit-learn · SQL · Power BI

Built an end-to-end predictive maintenance solution for IoT sensor data (100K+ readings). Trained Random Forest, Gradient Boosting, and LSTM models to forecast equipment failures and integrated the models with a Flask API for real-time alerts. Designed an interactive Power BI dashboard to visualize sensor trends and failure predictions.
Tech Stack: Python · Power BI · Flask · SQL

Built an interactive Power BI dashboard to visualize customer segmentation and key performance metrics using SQL-integrated datasets.
Implemented data cleaning and transformation pipelines for 50K+ records, improving reporting efficiency by 15%.
Used DAX, Power Query, and SQL joins for advanced KPIs and dynamic filtering.

🧩 Tech Stack:
Power BI · SQL · Excel · Python (Pandas)

🎯 Outcomes:
Delivered real-time analytics dashboard for business reporting
Enhanced data visibility and decision-making for marketing teams