Where I've interned.5
- Hands-on ethical hacking: Kali Linux, John the Ripper, steganography, traffic analysis
- Developed Phishing Email Detection using Scikit-learn + NLP integrated with Gmail & CheckPhish APIs
- Threat classification, risk assessment, incident response in simulated attack scenarios
- Studied ISO 27001/27002, PCI-DSS, OWASP, SIEM, DDoS mitigation
- Automated ITSM workflows across Incident, Request, Change, Problem & Knowledge modules
- Designed Service Catalog items, Flow Designer workflows, Visual Task Boards
- Configured ACLs, roles, UI policies; earned ServiceNow University credentials
- Cleared the ServiceNow Certified System Administrator exam on the $300 voucher the program provided
- Developed AI-driven threat detection and LLM security systems aligned with SOC workflows
- Built platforms to analyze emails, URLs, QR codes, and logs with LLM API integration
- Implemented prompt security controls — PII redaction, injection detection, RBAC, rate limiting
- Hands-on OWASP Top 10, SIEM, Snort IDS, Active Directory attacks & HIPAA compliance
- Built full-stack freelancing marketplace — project posting, proposals, real-time messaging
- JWT + Bcrypt auth, Socket.io live chat, end-to-end API design with Express
- Completed MongoDB Node.js Developer Path certification during internship
- Scripted a bash monitor to log CPU, memory & disk usage, scheduled via cron every 5 minutes
- Ran an app as a custom systemd service behind an Nginx reverse proxy, hardened with UFW (ports 80 & 22 only)
- Built automated infrastructure — scripted provisioning, centralized logging, and kill-and-recover resilience testing
- Recognized as a Featured Outstanding Intern — Summer Internship 2026 (July Cohort)
Things I've built.
PathFinder AI — Evidence-Based Adaptive Learning Path Engine
3rd Place · HCLTech AMPlified Season 1 · Team Lead, Team PERCEPTRON — an evidence-based learning engine that verifies skills from proof, not self-reports. • Builds a personal learning path from demonstrated evidence — GitHub, contest stats, real projects • 211-skill prerequisite graph, scheduled so quick wins come first • Fine-tuned retrieval model lifted Recall@5 from 56.9% → 93.8%, shipped as a 32MB on-device model • Nexus, the built-in AI mentor, guides with Socratic hints instead of answers
AI Email Threat Detection & Forensic Intelligence
Smart India Hackathon 2026 platform (AICTE problem statement) that detects phishing, BEC & malware emails and traces the sender. • Adaptive ML ensemble (XGBoost + LightGBM + DistilBERT) trained on 30k+ emails • Fuses VirusTotal, SafeBrowsing, PhishTank & AbuseIPDB into one 0–100 risk score • SPF/DKIM/DMARC header forensics, live threat map, Gmail inbox scan & .eml upload • URL-guard browser extension with click-to-scan and auto-block
SOC Log Analyzer — Detection & Incident Response Engine
Log-driven SOC analysis engine, split out of the earlier threat-detection platform into a standalone service. • Maps events to MITRE ATT&CK techniques and correlates IOCs across pipelines • Monitoring rules with live ingest, alerting and incident response playbooks • FastAPI surface for analyze / history / MITRE mapping / incidents — 31-test suite, CI green
LLM Guardian — Security Firewall for LLM Apps & AI Agents
Unified guardrail layer for LLM applications and agentic systems — the flagship that absorbed three earlier security builds. • 10 detectors covering prompt injection, PII, toxicity, URL threats, keyword & length policy • Verdict engine — allow / sanitize / quarantine / block — with Redis telemetry and a live SOC console • MCP tool-call security: tool poisoning, data leaks and multi-tool correlation; JWT + RBAC admin API • Compliance modes, safe-prompt cache and audit-ready reporting
AI-Kavach — Cyber Defense System
Co-developed, IEEE-presented platform defending users against modern web vulnerabilities. • Detects malicious phishing sites, malware, and anomalies • Implements client extension and centralized cloud telemetry • Triggers threat alerts and automated response actions
What I work with.
From the SOC floor to CTF rooms — detection, response, and offensive testing.
Ensembles, retrieval and agent guardrails — models on real request paths.
APIs, dashboards, and shipping full products end to end.
Where the builds actually run — deployed, not just local.
What I've earned.
Certifications
Where I've competed.
Led Team PERCEPTRON — a 5-member squad — as Team Lead through HCLTech AMPlified: The AI Challenge Season 1, coordinating the evidence pipeline, graph scheduling, and ML model work across the team; shortlisted Top 25 of 4,839 registered students across 97 partner colleges, then pitched PathFinder AI before the HCLTech panel in Round 2 and placed 3rd nationally. Built the evidence-based adaptive learning path engine: Noisy-OR evidence model over GitHub/LeetCode/resume signals, a 211-skill DAG scheduled with Kahn's algorithm, and a fine-tuned bi-encoder lifting Recall@5 from 56.9% to 93.8%.
Recognized as one of the Outstanding Interns in the Learnix Labs Summer Internship 2026 (July Cohort) for technical quality, consistency, and execution — automated workflows, managed infrastructure, and implemented efficient DevOps practices on real-world projects.
Team-built MCP Guardian — a real-time security layer for AI agents that detects and blocks prompt injection, tool poisoning, malicious documents, and sensitive data leakage before they reach the model or external tool. I owned the backend detection pipeline, API layer, and authentication system.
Reached Top 5% and Platinum League on TryHackMe with a 44-day streak across OSINT, malware analysis, Active Directory attacks, and penetration testing rooms.
Papers I've written.
First-author framework that catches hallucinations claim-by-claim in RAG systems: decomposes generated answers into atomic factual claims and verifies each against the retrieved evidence using a weighted fusion of semantic similarity, entity overlap, and NLI — surfacing Supported / Unsupported verdicts with explanations through a live auditing dashboard.
Analyzed QRishing attack vectors, proposed ML-based detection and mitigation strategies with defensive measures for QR code exploitation.
Developed middleware for real-time prompt compliance, PII redaction, toxicity detection, and responsible AI output validation.