AI Security × Full-Stack Engineering
SwatiArya
I build systems that hold up under adversaries:
forensic tooling, LLM security layers, tamper-evident infrastructure.
- 5+ yrs shipping
- 14 papers
- 4 hackathon podiums
FIG. 01 / PROFILEAI SECURITY · EST. 202001 — Profile
About
I am a full-stack developer and cybersecurity researcher who builds production systems where correctness is not optional: fraud-detection graph engines, tamper-evident logging, LLM security layers.
Currently a Software Engineering Intern at I4C, Ministry of Home Affairs, building tooling for live cyber investigations and OSINT.
14 peer-reviewed papers across Scopus / WoS / SCI. MCA in Data Science & AI. I bridge engineering execution and the research that backs it.
AI SECURITY / ENGINEER
SECURE · BUILD · RESEARCH
SPEC.001
FOCUS : AI SECURITY
STACK : PY / TS / JAVA
DOMAIN : CYBER / FORENSICS
OUTPUT : PROD SYSTEMS
RESEARCH: 14 PAPERS
VER. 2.0.0
DATE. 2026
ID // SWATI-ARYA
CATEGORY
SECURITY
FULL-STACK
FOCUS
· AI GUARDRAILS
· FORENSIC GRAPHS
· TAMPER-EVIDENCE
Presence & credentials
LIVE / VERIFIEDCurrent · Core Team
At LV8
AI Security Lead · Full-Stack Dev · Business Analyst
At LV8 I work across the whole stack and the whole business: I scope and build the AI security layers, ship full-stack product, and do the strategic work behind it, client proposals, pricing analysis and competitive research.
I bridge engineering execution and the decisions that back it, so the thing we ship and the case we make for it stay in sync.
- 01
AI Security
LLM guardrail Layer 1 (open-source), RAG pipeline security, security architecture across client builds.
- 02
Full-Stack Delivery
AI calling service, deployed web prototypes, and the Elevate Health patient dashboard visual tracker.
- 03
Business Analysis
Client proposals, tiered pricing models, hosting/token cost research, competitor and CSO-level analysis.
Capabilities
AI Security
- LLM guardrails
- RAG security
- AI red-teaming
- Prompt hardening
Cyber / Forensics
- OSINT
- Fraud graphs
- Hash-chaining
- RBAC / least-priv
Full-Stack
- TypeScript
- React / Next.js
- Python / FastAPI
- Node · SQL · Mongo
Research
- Threat modeling
- Dataset curation
- Reproducibility
- Academic writing
Doctrine
Correctnessis notoptional.
Security is not a feature you bolt on at the end. It is the shape of the system from the first commit: tamper-evident by default, adversary-aware by design, verifiable on demand.
- 5+Years shipping production systems
- 14Peer-reviewed papers · Scopus / WoS / SCI
- 4National hackathon podiums
- ~140Google Scholar citations
02 — Peer-reviewed
Research
03 — Selected builds
Work
Production systems where correctness is not optional. Forensic graphs, tamper-evident infrastructure, security testing.
01Money-Mule Detection Graph Engine
Transaction-graph engine detecting cycles (3–5 hops), smurfing (fan-in/out ≥10 in 72h) and layered shell chains; a GenAI layer writes plain-language fraud-ring explanations, output as evidence-tagged JSON for case files.
→02Audit-Ready Logging & Evidence Metadata
Append-only, tamper-evident hash-chained logs with retention controls, on-demand cryptographic verification and automatic PII redaction. Built for chain-of-custody.
→03Security Invariants Runner
An OpenAPI-driven runner that continuously checks baseline API security: auth on writes, admin-route protection, sensitive-field scanning, CORS and rate limits. Fails loud on drift.
→04Digital Footprint Vault
Aggregates a person's online presence, flags breaches and gives cleanup guidance. Placed 5th nationally at HackSetu 1.0.
→- Money-mule v2
- LLM Layer-1
- Invariants CI
Let’s build something secure.
AI security, forensic tooling, or full-stack systems that have to be right.
Open to collaborate · Lucknow / New Delhi, India