Ansh Singhal, AI/ML EngineerHome
Hire me
Open to opportunities

AnshSinghal

I build AI security gateways.

Final-year computer science student at Bennett University and AI/ML engineer at CyberUltron, where I architected and own the AI Mesh Firewall, an AI security gateway handling 100k+ requests a second. I also lead ZeroDayShield, an independent zero-day research engine, and contribute to open-source threat intelligence.

Greater Noida, IndiaB.Tech CSE (AI), class of 2027
100k+requests a second through the AI gateway I built
1,000+enterprise endpoints running AIGuardX
4production analyzers shipped to IntelOwl
9.47CGPA at Bennett University

About me

I'm an AI/ML engineer and a final-year B.Tech student in Computer Science with an AI specialization at Bennett University, holding a 9.47 CGPA.

Since February 2026 I've worked remotely at CyberUltron, where I own the AI Mesh Firewall end to end, shipped AI-powered remediation on an application security platform, and built core features of AIGuardX. Before that I spent 16 months contributing security analyzers to IntelOwl.

Outside work I co-organize GDG Bennett, sit on Rotary International's Youth Advisory Council as its only representative from Asia, and help student founders through the E-Cell. Off the clock, I'll happily defend a Harry Potter fan theory. There's one on the card.

CurrentlyAI/ML Engineer at CyberUltron
StudyingB.Tech CSE (AI), Bennett University, 2023 to 2027
FocusAI security, agentic RAG, real-time systems
Based inGreater Noida, India
Fun fact, straight from my GitHubMaybe Dumbledore used Harry Potter as a weapon to kill Voldemort.

Experience

Download full resume

AI/ML Engineer

CyberUltron Consulting Pvt. Ltd.
CurrentFeb 2026 to presentRemote
AI Mesh Firewall

Architected and shipped a production AI security gateway as sole owner across backend, frontend, DevOps, cloud deployment and scaling. It enforces centralized pre- and post-LLM security, PII detection and prompt-injection mitigation.

100k+ RPS500+ endpoints−99% unsafe LLM behavior
Application security platform

Drove critical feature expansion: added CBOM scanning coverage, shipped AI-powered vulnerability remediations and optimized platform performance.

−60% manual triage time+40% performance
AIGuardX

Engineered core features of an endpoint AI governance product that stops sensitive data leaking through widely used enterprise AI tools, using a lightweight agent and a local proxy.

1,000+ endpoints5+ AI tools covered

Open Source Developer

IntelOwl Threat Intelligence Platform
Jan 2025 to Apr 2026Remote
+23%threat detection coverage from 4 production security analyzers
99.9%migration success after fixing Django migration and data consistency issues
pytestvalidation suites for analyzer reliability and regression protection
Dockernetworking and caching fixes for stable deployments and analyzer runs

B.Tech in Computer Science (AI)

Bennett University, The Times Group
CGPA 9.47Aug 2023 to May 2027Greater Noida

Projects

Every card is live. Poke at them.

Featured

AI Mesh Firewall

A production AI security gateway I own end to end, across backend, frontend, DevOps, cloud and scaling. It fronts 500+ endpoints, enforces pre- and post-LLM security, and scales with async concurrency from 100 to 100k+ RPS.

100k+RPS throughput
−99%unsafe LLM behavior

“Summarize the indemnity clause in this contract.”

AuthPassed
PIIPassed
InjectionPassed
LLMPassed
OutputPassed

Passed all five checks and reached the model.

ZeroDayShield

Open live site

An independent zero-day research engine I lead and build from scratch: 50+ custom static-analysis engines with zero open-source libraries, mining live registry traffic, plus a multi-stage LLM triage pipeline.

npm
49,522 signals cross-checked against 228,020 known-malicious packages
100%recall
93%precision
15×lower triage cost

An agentic RAG engine for legal research over 10,000+ documents, with citation-traced answers, multi-agent query decomposition, reranking and deterministic grounding on AWS microservices.

97%retrieval accuracy
<500msP95 latency
10,000+legal documents

VoiceOps AI

Code

A real-time voice-streaming backend holding sub-200ms latency across 100+ concurrent sessions: WebSocket ingest, Kafka consumer groups, and Kubernetes HPA that autoscales on consumer lag.

pod 1pod 2idleidleidleidle
Consumer lag
low
Autoscaling simulation: 2 consumer pods for this load. Real latency stays under 200ms.

Published research: a physics-aware GAN that translates all-weather SAR radar into optical-quality flood imagery at 1024×1024, with a dual-head generator that learns flood boundaries as water, not texture.

Sentinel-1 radar image of flooded farmland in SpainSentinel-1 radarSentinel-2 optical
Real radar and optical images of the same flooded farmland in Spain (Sen1Floods11 dataset). Flood-GAN learns to turn the left into the right. Contains modified Copernicus Sentinel data.

Skills

My periodic table. Hover an element to see where I used it, then drop three into the mixer and rebuild something I shipped.

0 of 5 projects rebuilt
56 elements, 9 groupsHover or tap an element to see where I used it. Each click also drops it in the mixer.
Also in my projects

Element mixer

Pick up to four elements from the table. The right three rebuild one of my projects.

Nothing in the flask yet.
Rebuilt so far
AI Mesh FirewallZeroDayShieldDHARAVoiceOps AIIntelOwl
Optional side quest

Try to break my AI

Mesh guards a passphrase behind six layers, each one a defense I ship in production. Nothing on this site is locked. This is just for fun.

Layer 1: Bare modelOnly a system prompt asks the model to keep the passphrase.

I am Mesh. I guard a passphrase. Try me.

Mesh is a real model behind real guardrails. You get 20 tries a day.

Community and leadership

4,000+GDG Bennett UniversityCo-organizer. Led 12+ events for 4,000+ participants and grew the chapter by 40%.
15+Entrepreneurship CellJoint Secretary. Supported the launch of 15+ student startups.
AsiaRotary InternationalYouth Advisory Council. The only representative from Asia on global youth initiatives.

Leave a note

Notes pass through the same checks as my firewall before they post. Emails and phone numbers get redacted, injections get stopped.

Drag the globe to spin it

Who's been here

0visitors so far
0today
1here right now
0countries
0cities
LiveWaiting for the first visit…