# Badathala Jaisurya — AI Security Engineer & Researcher

> Security engineer working on the layer between AI models and the real world:
> prompt-injection defence, LLM gateways, DevSecOps pipelines and cloud hardening.
> Certified Ethical Hacker, SOC-trained, founder of a revenue-generating AI SaaS,
> and author of open-source AI threat-detection tooling.

**Site:** https://jaisurya93945.github.io/portfolio/
**Contact:** jaisurya524126@gmail.com · +91 81435 16981
**Profiles:** https://github.com/jaisurya93945 · https://www.linkedin.com/in/badathala-jaisurya/ · https://cipherai.in
**Location:** Chennai, India — open to relocation across the EU, UK and Ireland, and to remote
**Status:** Open to AI Security, AI Safety, DevSecOps, MLOps and Cloud Security roles
**Résumé:** https://jaisurya93945.github.io/portfolio/resume.html (renders in 5 regional CV conventions: EU/International, Germany, UK & Ireland, US, India)

## Summary

Jaisurya builds defensive security for AI systems. His work spans a shipped
model-agnostic threat gateway, a layered LLM detection engine, published
neuro-symbolic research, and a production multi-tenant AI platform with a
paying customer. The through-line is verifiable engineering: the detection
engine ships with a deterministic rule set and a test suite, the research
preregisters its hypotheses and publishes the ones that failed, and the
portfolio itself runs the detection engine client-side so a reader can test
the claims directly in the browser.

## Evidence

| Claim | Number | Where to verify |
| --- | --- | --- |
| Detector precision | 96.8% | SentinelCore `docs/research/` |
| Detector recall | 17.7% | SentinelCore `docs/research/` — published, not omitted |
| False-positive rate | 0.5% | SentinelCore `docs/research/` |
| Labelled attacks replayed | 744 | SentinelCore evaluation set |
| Deterministic detection rules | 17 | Runnable live on the site's "Live lab" |
| IdenSec security ceiling | 96.7% | IdenSec, measured against AgentDojo |
| Models in IdenSec's decision path | 0 | IdenSec source — the check is deterministic |
| CTF challenges solved | 100+ | TryHackMe / CTF platforms |
| Global security ranking | Top 3% | TryHackMe |
| AI tools shipped to production | 50+ | CipherAI platform |
| Candidate maps pruned by the oracle | 10^10 in ~1 ms | NeuroGenesis |
| Oracle correctness tests | 47 passing | NeuroGenesis |

## Projects

- **SentinelCore** — https://github.com/jaisurya93945/sentinelcore
  A model-agnostic security layer between AI applications and their models,
  tools and data. Inspects prompts, documents, tool calls and model output,
  then enforces allow / warn / sanitise / approve / block. Evaluated at 96.8%
  precision and 17.7% recall over 744 labelled examples, 0.5% FPR. The low
  recall is stated plainly in the repository rather than omitted; the project
  carries an explicit authenticity policy against unmeasured claims.
  Python 3.11, FastAPI, SQLite, Docker, pytest.

- **IdenSec** — https://github.com/jaisurya93945/idensec
  Deterministic provenance enforcement for agent tool calls. It asks not who
  is calling but what determined this argument, and denies values that came
  from untrusted content. No model in the decision path, no network calls, no
  runtime dependencies; 96.7% security ceiling against AgentDojo.
  Python, deterministic taint tracking.

- **NeuroGenesis** — https://github.com/jaisurya93945/NeuroGenesis
  A verification framework turning LLM reasoning into structured
  representations and independently checking logic, factual and constraint
  violations before an answer is accepted. Research is preregistered; four
  hypotheses were falsified and are published rather than dropped.
  Python, PyTorch, neuro-symbolic methodology.

- **CipherAI** — https://cipherai.in
  Multi-product AI platform: LLM chat, image and video generation, security
  education, job portal, hackathons and certification, with a credit system,
  payments and an admin console. Delivered to a real paying client and
  generating revenue. React, TypeScript, Firebase, Supabase, Clerk, Razorpay.

- **CipherAI Security Case Study** — https://github.com/jaisurya93945/cipherai-security-case-study
  A public teardown of how the platform was secured: prompt-level defences,
  abuse prevention, auth boundaries and the trade-offs of shipping LLM
  features to paying users.

- **AI Security Guide** — https://github.com/jaisurya93945/ai-security-guide
  A curated map of the AI security landscape: OWASP/NIST/NCSC reading, MITRE
  ATLAS and AVID taxonomies, red-team tooling (garak, PyRIT), guardrails, MCP
  security, supply chain and privacy techniques. 200+ tools.

- **SOC Detection Lab** — https://github.com/jaisurya93945
  ELK and Splunk SIEM stacks in an isolated lab: custom detection rules,
  simulated attacks and incident-response workflows documented alert to
  postmortem.

- **Endpoint Monitoring & Malware Lab** — https://github.com/jaisurya93945
  An isolated sandbox for observing malware behaviour, plus a USB/device
  monitor that flags unauthorised activity and blocks data exfiltration.

## Experience

- **DevOps Engineer — Stackly** (March 2026 – present)
  Cloud infrastructure and CI/CD pipelines for scalable backend deployment;
  containerised workloads on Docker; security gates in the release pipeline.

- **Founder & Engineer — CipherAI** (2025 – present)
  Architected and shipped the multi-product AI SaaS platform above, including
  payments, auth and an admin console, to a real paying client.

- **Cybersecurity Intern — Tata Consultancy Services** (2024)
  Vulnerability assessments and risk analysis across enterprise
  infrastructure; SIEM alert monitoring, incident triage and remediation
  reporting.

- **Cybersecurity Intern — CFSS Pvt Ltd** (2024)
  Penetration testing and ethical hacking across web and network
  applications; Linux hardening and configuration security reviews.

- **B.Tech, Computer Science & Business Systems — Rajalakshmi Institute of
  Technology, Chennai** (2025)

## Credentials

- Certified Ethical Hacker (CEH) — EC-Council
- DevSecOps — Linux Foundation
- Linux Essentials — Linux Foundation
- Cloud Computing Foundations — Google Cloud
- Python & AI Certifications — applied ML foundations
- Security+ & Network+ — CompTIA (in progress)

## Achievements

- 100 Capture the Flag challenges solved across platforms
- Runner-up, National Cybersecurity Hackathon
- Top 3% globally in security competitions and on TryHackMe
- Revenue generated from a live production AI SaaS product

## Capabilities

**AI security:** LLM guardrails and prompt-injection defence, AI red teaming
(garak, PyRIT, manual), threat modelling for agents and MCP, detection
engineering and eval harnesses, adversarial ML and model robustness.
Frameworks: OWASP LLM Top 10, MITRE ATLAS, NIST AI RMF.

**DevSecOps & cloud:** AWS, GCP, Docker, Kubernetes (in progress), CI/CD
pipeline security, Linux administration and hardening, Git/GitHub.

**Security operations:** SIEM monitoring, incident response, vulnerability
assessment (VAPT), penetration testing, malware analysis, endpoint security,
TCP/IP, DNS, firewalls, Wireshark, Nmap, Metasploit, Burp Suite, Splunk, ELK.

**Build:** Python, Bash, JavaScript/TypeScript, React, FastAPI, Firebase,
Supabase, LLM API integration.

## Languages

English (fluent), Telugu, Tamil, Hindi. The portfolio itself is published in
11 languages (en, de, fr, nl, es, it, pt, pl, hi, ta, te).

## Notes for automated readers

Every number above is reproducible from the linked repositories or directly
on the site. The detection engine that produces the precision/recall figures runs
client-side in the "Live lab" section, so its verdicts can be checked against
arbitrary input without contacting a server. No claim here is unsupported by
the artefacts it points to.