Projects

Preemptive DDoS Attack Forecasting

ResearchIn-Progress2026
  • Collaboration with Tata Communications Limited through Prof. Ganapathy Krishnamurthi, IIT Madras.
  • Developing a forecasting pipeline on NetFlow host features to predict volumetric DDoS attacks 1-3 minutes before attack onset; delivered preliminary results predicting 52% of attack episodes at ~11 alerts/day.
NetFlowForecastingDDoSNetwork Security

Dental CBCT Simulation

ResearchIn-Progress2026
  • Collaboration with C-DAC through Prof. Ganapathy Krishnamurthi, IIT Madras.
  • Developed a high-fidelity Monte Carlo simulation of a dental CBCT system using OpenGATE 10, generating cone-beam projections matched to the Mercu1012X flat-panel detector and RADII KL3SB0.5-130 X-ray source.
Monte CarloCBCTMedical Imaging Simulation

MedPAO: A Protocol-Driven Agent for Structuring Medical Reports

ResearchPublished2025
  • MedPAO is an intelligent agent system designed to process and structure medical radiology reports according to the standardized P.A.B.C.D.E.F protocol for chest x-ray interpretation.
  • The system extracts medical concepts from free-text radiology reports, maps them to standard medical ontologies, and organizes them into a structured clinical format. Paper accepted in MICCAI 2025
LLMAI AgentPAO LoopMedical grounding

CXR Agent- Comprehensive Chest X-Ray Analysis System

ResearchIn-Progress2024-2025
  • A state-of-the-art AI system that combines advanced computer vision techniques with agentic Retrieval-Augmented Generation (RAG) for comprehensive chest X-ray analysis.
  • Integrates multiple AI models and medical knowledge bases to provide detailed diagnostic insights.
LLMMCPRAGAI Agent

Efficient Knowledge Distillation of SAM for Medical Image Segmentation

ResearchPublished2025
  • Developed KD SAM which effectively balances segmentation accuracy and computational efficiency, making it well-suited for real-time medical image segmentation applications in resource-constrained environments.
  • Implements advanced knowledge distillation techniques to compress the Segment Anything Model (SAM) while maintaining high performance.
SegmentationKnowledge distillationPerceptual lossPrompt-based segmentation

Automated Non-Invasive Diagnosis of Diabetic Retinopathy using CNN

DesktopCompleted2023
  • Developed a portable, software-driven imaging system using Raspberry Pi to capture retinal fundus images and predict the severity of diabetic retinopathy using CNN.
  • Cost-effective solution for early detection of diabetic retinopathy in resource-limited settings.
CNNRaspberry PiClassificationImaging