These are some of my Artificial Intelligence, Machine Learning and Data Science Projects. More of them can be seen on my GitHub account

December 9, 2024

Project-1 : Driver Behavior Analytics System

Aim: Built comprehensive driver risk analytics system using survival analysis and Bayesian modeling for 300K+ drivers with real-time assessment capabilities.

Key Metrics: 88% churn prediction accuracy, 40% reduced claim processing time, 10K+ daily API requests with < 200ms response time.

Tech Stack: Apache Kafka, AWS S3, PyMC3, FastAPI, PostgreSQL. Integrated Bridgestone ecosystem data with Cox regression and MCMC sampling for production deployment.

May 15, 2025

Project-2 : Production LLM Serving Optimization Framework

Aim: High-performance LLM serving framework with vLLM continuous batching, INT8/INT4 quantization, and multi-GPU tensor parallelism for enterprise AI applications.

Key Metrics: 12.3K+ requests/sec throughput, 42ms P50 latency, 70% memory reduction, 1500+ concurrent users, 2.3x speedup with custom CUDA kernels.

Tech Stack: vLLM, FastAPI, Docker/Kubernetes, Prometheus/Grafana. Implemented distributed inference with token streaming and production monitoring.

November 9, 2023

Project-3: Face-Mask-Detection

Aim: Real-time face mask detection system using Deep CNN with transfer learning for automated safety compliance monitoring in public spaces.

Key Metrics: 99.60% accuracy on 12K+ images, < 50ms inference time, 500+ concurrent users with VGG16 transfer learning architecture.

Tech Stack: TensorFlow, VGG16, Flask, HTML/CSS/JS. Applied data augmentation and custom CNN layers for production web deployment.

October 9, 2023

Project-4: Customer Churn Prediction

Aim: Predictive analytics system for telecom customer churn with explainable AI insights and real-time risk scoring for proactive retention strategies.

Key Metrics: 92% churn prediction accuracy, 100K+ customer records processed, 95% precision for high-risk identification with SMOTE+ENN balancing.

Tech Stack: Random Forest, SMOTE+ENN, Flask, Pandas. Implemented RFM analysis and behavioral pattern detection for telecom domain.