Helping organisations break down real attack surfaces across Web, Mobile (Android/iOS), API, and now AI/LLM systems — through deep manual penetration testing aligned to OWASP, MASVS, PCI DSS 4.0, and OWASP Top 10 for LLMs.
Senior Security Analyst specialising in penetration testing for Web, Mobile and API platforms — now actively expanding into AI/LLM adversarial testing. I combine offensive security depth with a clear understanding of business context, helping leadership understand the true risk behind each finding — not just a CVSS number.
I embed seamlessly into engineering, QA and architecture workflows, treating security as a continuous SDLC checkpoint rather than a final-sprint gate.
vasu@secops $ run-engagement --scope web,api,mobile,ai --framework OWASP,MASVS,PCI-DSS,LLM-TOP10
[✓] Threat-driven test plan initialised — manual-first methodology [✓] False-positive separation layer: enabled [✓] Business impact mapping: active [+] Loading AI/LLM adversarial modules... [✓] Prompt injection scanner: ready [✓] Model jailbreak probe set: loaded (47 vectors) [✓] RAG poisoning detection: enabled [✓] Report: executive-ready + developer-ready + compliance-ready [✓] Retest & closure validation: includedvasu@secops $ _
Deep manual testing for authentication flaws, broken access control, IDOR, session weaknesses, business logic gaps, and OWASP Top 10 coverage before every production rollout.
MASVS-aligned assessments for Android & iOS — data storage, API usage, reverse engineering resistance, certificate pinning, and runtime protection validation.
BOLA, BFLA, JWT weaknesses, rate-limit bypass, mass assignment, replay attacks, and abuse scenarios across microservices, REST, GraphQL, and API gateways.
Adversarial assessment of LLM-powered products: prompt injection, jailbreaks, RAG poisoning, data exfiltration via model outputs, agent trust boundaries, and OWASP LLM Top 10 coverage.
SAST, SCA and secrets scanning integration into CI/CD with meaningful quality gates, low noise signal-to-ratio tuning, and training for developer security ownership.
IAM misconfiguration, storage exposure, network perimeter, and internal service discovery checks across AWS and Azure environments with actionable hardening guides.
As AI-powered products go mainstream, so do their attack surfaces. I'm actively building expertise in adversarial AI testing — probing LLMs, RAG pipelines, AI agents, and model APIs for vulnerabilities that traditional pentesting doesn't cover.
Grounded in the OWASP Top 10 for LLMs and MITRE ATLAS, my approach treats AI components as first-class attack targets with unique threat models.
Direct and indirect prompt injection attacks that manipulate model behaviour, bypass system prompts, or hijack agent actions through untrusted inputs.
Multi-turn roleplay attacks, DAN variants, adversarial suffixes, and encoding tricks that strip guardrails from production LLMs.
Injecting malicious content into retrieval corpora, vector stores, or knowledge bases to poison model responses at inference time.
Extracting training data, system prompt leakage, and sensitive context exfiltration through crafted adversarial queries and model inversion techniques.
Exploiting autonomous AI agents to take unintended actions — file access, API calls, privilege escalation through indirect prompt injection chains.
vasu@secops $ probe-llm --vectors prompt-injection,jailbreak,rag-poison,exfil
[+] Scanning system prompt boundaries... [!] LLM01 — Prompt Injection: VULNERABLE — indirect via document upload [!] LLM02 — Insecure Output Handling: PARTIAL — XSS via rendered markdown [!] LLM06 — Sensitive Info Disclosure: VULNERABLE — system prompt leakable in 3 turns [✓] LLM04 — Model DoS: Not exploitable — rate limiting enforced [+] Running jailbreak probe set... [!] Safety bypass: SUCCESSFUL via multi-role persona chain attack [AI] Generating adversarial remediation roadmap... [✓] Report: OWASP LLM Top 10 mapped, business impact rated, fix guidance includedvasu@secops $ _
Full-stack web pentest for a payment-handling portal. Uncovered broken access control chains, insecure session management, and IDOR across user account flows — all before production launch.
End-to-end MASVS and OWASP API Top 10 assessment covering data-at-rest protection, JWT weaknesses, certificate pinning bypass, and BOLA across supporting microservices.
Web and mobile testing mapped to PCI DSS 4.0 requirements for a payment-focused product team, enabling smoother QSA audit review and faster compliance sign-off.
Adversarial testing of a customer-facing AI assistant. Discovered prompt injection via user-controlled document uploads leading to full system prompt disclosure and safety guardrail bypass.
Security review of an enterprise RAG deployment. Identified knowledge-base poisoning vectors via unsanitised document ingestion and model output injection enabling cross-user data leakage.
IAM privilege analysis, S3 public access audit, internal service exposure mapping, and metadata API hardening for a SaaS platform migrating workloads to AWS.
Need a focused pentest for an upcoming release, an AI/LLM security assessment, PCI DSS-aligned testing, or an independent review of an existing application? Reach out — I'll respond within one business day.