Projects
🔐 CopyGuard – Serverless AI Code Detection Platform
Tech Stack: AWS Lambda, Bedrock Claude v2, Terraform, S3, CloudFront, Grafana, MLOps
GitHub Repo | Blog
- Built a production-grade platform to detect AI-generated code using Amazon Bedrock (Claude v2).
- Achieved ~99.9% availability with sub-2s response time via AWS Lambda.
- Deployed complete IaC stack with Terraform (API Gateway, IAM, logging, CORS).
- Monitored system latency and model confidence using Grafana dashboards.
- Outputs versioned and stored in S3 for audit and reproducibility.
🔐 ThreatMatrix – MLOps Pipeline for Network Intrusion Detection
Tech Stack: Python, FastAPI, MongoDB, MLflow, DagsHub, Docker, GitHub Actions, AWS EC2
GitHub Repo | Blog
- Developed an end-to-end MLOps system for real-time intrusion detection.
- Modular pipeline with custom internal Python package for data ingestion, validation, transformation, training, and prediction.
- Containerized pipeline using Docker and CI/CD via GitHub Actions; images deployed to Amazon ECR.
- Real-time FastAPI endpoints (
/train,/predict) served on AWS EC2 with <15ms latency. - Tracked experiments and metrics using MLflow + DagsHub.
🏥 MP4-to-DICOM Conversion Pipeline (For AIIMS Delhi)
Tech Stack: In progress (under evaluation)
No public code or blog available yet
- Currently contributing to a project under the guidance of Prof. Dr. Deepak Agrawal (AIIMS Delhi), aimed at automating the conversion of MP4 videos to DICOM format for radiological workflows.
- The project is designed to benefit radiology departments at AIIMS, aligning with broader clinical infrastructure goals.
- Focused on pipeline architecture, tool selection, and reproducibility.
- Currently in the planning and prototyping phase.
🌿 Tomato Leaf Disease Detection
Tech Stack: TensorFlow, Keras, Xception, Seaborn, Matplotlib
GitHub Repo | Blog
- Achieved 97.23% accuracy using a fine-tuned Xception model with hyperparameter tuning.
- Used data augmentation, transfer learning, and dropout regularization.
- Backed by an IEEE conference publication.
🧠 Breast Cancer Detection Using Deep Learning
Tech Stack: CNNs, Population-Based Training, Transfer Learning
GitHub Repo | Blog
- Developed a deep learning pipeline for breast tumor classification from mammogram images.
- Achieved 98.46% accuracy using Population-Based Training and model fine-tuning.
- Work presented at an IEEE conference and now extended to uncertainty-based model calibration for journal submission.
For full research details and metrics, visit my Publications or Experience pages.