Dinesh Karthick B.DineshKarthick B.

A CPU-only Windows PE malware-screening system that separates invariant structural screening from mutable full-file context analysis.

• Two independently trained models, a 73-feature structural Anchor model and a 2,568-feature full-context model, joined by an asymmetric decision policy.
• Three explicit outcomes, MALICIOUS_SCREEN, CLEAN_SCREEN and REVIEW, with fail-closed handling for uncertainty, malformed inputs, timeouts and resource limits.
• Order-statistic Neyman-Pearson calibration and transformation-contract testing to control benign false alarms and keep scores stable under permitted edits.
• A bounded scanner daemon with a 32 MiB input limit, a 250 ms scan deadline and zero waiting queue.

An offline tool that assesses non-human identity exposure and how far a leaked credential could reach in CI/CD environments.

• Analyses GitHub Actions workflows, Terraform plan and state inputs, and IAM policy JSON to reconstruct potential authority paths.
• Classifies revocation survival as R (ends with the entry point), E (expires automatically) and Z (survives) to derive a rotation order.
• Generates evidence-backed HTML, JSON and SARIF reports without executing scanned content or making live cloud or network calls.

A Python pipeline for processing environmental DNA datasets and supporting biodiversity assessment.

• Worked on sequence processing, clustering and embeddings for environmental DNA data.
• Explored interactive analytical visualisations built with Streamlit.