Developed a centralized platform to detect and coordinate the takedown of phishing domains targeting public sector banks. The solution integrates AI-based phishing detection, real-time reporting, and global registrar coordination to streamline cross-jurisdictional takedown efforts and reduce response time.

Built a programmable safety framework that sits between users and Large Language Models (LLMs) to enforce policy, detect jailbreak attempts, prevent sensitive data leakage, and log auditable safety decisions. Core features: prompt/response risk inspection, jailbreak detection, sanitization controls, automated LLM red-team simulator, bypass rate measurement, latency tracking, safety scorecard generation.

A privacy-preserving cybersecurity platform built to counter AI-driven threats including deepfakes, phishing automation, and data leakage. Built on a React.js and Django stack, it integrates forensic diagnostics, adversarial simulation, and active defense into a single ecosystem. Its core innovation — a Compression-Resistant Adversarial Cloak — generates platform-specific noise that survives social media compression to protect biometric identity. A Zero-Knowledge Adversarial Threat Twin (Z-KATT) engine safely simulates realistic cyberattacks using synthetic, legally compliant data, performing adversarial manipulation, forensic delta analysis, and risk scoring without touching real user data.

Banks accumulate thousands of forgotten API endpoints over years of digital transformation. Revenant scans across 6 discovery vectors — network, API gateways, code repositories, CI/CD pipelines, and traffic logs — to build a complete API inventory, classifying every endpoint as Active, Deprecated, Orphaned, or Zombie. Confirmed zombie APIs are routed through an automated decommissioning workflow covering owner identification, gateway blocking, and repository cleanup. A built-in Risk Correlator maps APIs against the OWASP API Top 10, runs CVE lookups, detects PII exposure, and assesses regulatory impact against RBI, PCI-DSS, and DPDP Act 2023.

Designed a multi-layered, zero-trust cybersecurity architecture to protect CCTV networks against hacking and data breaches. Secured the full pipeline from IP camera to NVR using VPN tunnels, TLS encryption, and RBAC with continuous identity verification. A network-based IDS/IPS monitored traffic using signature- and anomaly-based detection, while threat intelligence feeds dynamically updated detection rules. Attack vectors including RTSP hijacking, credential brute-forcing, and MITM attacks were simulated in controlled environments to validate detection and response capabilities end-to-end.

Developed a multi-utility cybersecurity toolkit that integrates several common offensive and defensive features into a single interface — functioning as a "Swiss Army Knife" for security testing, reconnaissance, and quick incident response.

Built a personalized book recommendation system that suggests titles based on user preferences and book similarities. The model uses both collaborative and content-based filtering to enhance recommendation accuracy and user engagement.