uptime: ai security × deepmind · mountain view

Ayush
Khandelwal

Software engineer securing frontier AI systems: adversarial ML defenses, prompt-injection red teaming, and model access control at Google scale. I've protected Gemini model weights, shipped integrity classifiers across billions of Search queries, and built threat detection at IBM. Published in PNAS & Nature. TEDx speaker.

view work get in touch
Ayush Khandelwal
./assets/photo.jpg

~/experience $ tail -f career.log

Work

→ present · mountain view

Software Engineer, AI Search Security with DeepMind

Google

  • Lead engineer for multimodal RAG poisoning mitigations on Search, covering text & video injection vectors across AI Overviews and AI Mode
  • Shipped high-precision, low-latency classifiers that suppress adversarially self-promotional and biased content in AI Mode, improving result integrity & factuality across billions of queries per month
  • Architected self-improving red-teaming systems that autonomously generate, mutate, and score prompt-injection attacks, surfacing novel attack classes into Gemini post-training and Cyber model training
  • Designing RL / continual-learning environments so RAG poisoning defenses adapt as attack strategies shift, in collaboration with DeepMind

→ · sunnyvale

Software Engineer, Core Security

Google

  • Secured access to Gemini model weights by building automated observability and remediation across model surfaces, closing over-provisioned access paths
  • Engineered an end-to-end system streamlining identity management on cloud clusters worldwide, substantially reducing over-provisioning threat vectors
  • Tech lead for 4 interns shipping the AI Access Remediation Engine & Agentic Access Platform, shifting from reactive reviews to automated remediation

→ · cambridge

Staff Software Engineer, Security Innovation Team

IBM

  • Led technical development of a fault-tolerant security alerting system serving 350k employees
  • Implemented NLP models for data anonymization on IBM Guardium, protecting sensitive data in logs

→ · cambridge & new york

Software Engineer Intern ×2

IBM

  • Built a real-time distributed threat detection engine processing 6M+ logs/day
  • Built an ETL pipeline covering 170k+ employees for ML-based insider threat detection on the Elastic Stack

~/projects $ ls -la --sort=impact

Projects

GroundTruth AI

gemini hackathon winner

Agentic fact-checking for YouTube: a swarm of agents validates news claims and balances perspective in real time. Winner, Google Gemini Hackathon (Jan 2025).

gemini agents fact-checking

NeuroLearn

ted ai winner

Led a team of 5 building a personalized education platform that turns EEG signals into engagement-based knowledge graphs, paired with AI voice generation. Winner, TED AI Multi-Modal Hackathon (Oct 2023).

eeg ml edtech
devpost

MarvinGPT

openai finalist

Led design and technical development with a team of 3 on a ChatGPT plugin that reads facial features and tone to give the model contextual understanding of a user's emotional state. Finalist, OpenAI Hackathon (May 2023).

llm multimodal chatgpt-plugin
devpost

IBM BroadBeach

$1M funded

Engineered and led design for a cloud security management tool automating security posture analysis and remediation, which received over $1M in funding at IBM and won the Rhonda Childress Award.

cloud-security automation

SixthSense

meta featured

Interact with authors as if they're beside you. Neurofeedback signals adapt your reading in real time. Meta Llama 3 Hackathon finalist, featured on Meta's blog.

bci llama3 nlp
devpost

OneFeed

AI-curated timeline unifying X, Instagram, and YouTube into a single ranked feed, surfacing what matters across your social graph.

ml react apis
devpost

~/awards $ ls | head -7

Recognition

Core Tech Impact Award Google
Winner, Google Gemini Hackathon GroundTruth AI
Winner, GHacks '24 Led team of 4 · neurofeedback AI tutors, with GDM
Gold Prize, Google Security & AI Hackathon Led team of 5 · agentic ACL generation engine
Winner, TED AI Multi-Modal Hackathon NeuroLearn · San Francisco
International ACM Student Excellence Award As Chair, UMass ACM
Rhonda Childress Award for Inventive Design IBM · BroadBeach cloud security

~/talks $ mpv --list featured/*

Talks

Stanford AI Summit: Future of Agentic Security at Google
Stanford AI Summit

Future of Agentic Security at Google

· w/ Heather Adkins · Palo Alto
TEDx Boston: NFTs are more than Art
TEDx Boston

NFTs are more than Art

· 9 min · Boston
talk

Embodied Intelligence: Physical AGI

MIT × Google AI Summit · Mountain View ·

talk

Future of Security Orchestration Automation Response

A16Z Tech Week · San Francisco ·

panel

Future of AI Hardware, with the CTO of Groq

MIT Imagination in Action · Cambridge ·

talk

Multimodal AI

MIT Imagination in Action · Cambridge ·

talk

NeuroLearn and the Future of AI-based Education

TED AI Startup Stage · San Francisco ·

panel

Security and Privacy Panel

Imagination in Action @ MIT Media Lab ·

panel

Bitcoin Research and Development Panel

MIT Bitcoin Expo @ CSAIL ·

judge

GHacks '25 & '26 · Technical Mentor, AI Security at Google for Startups

Judge & mentor · →

~/research $ grep -r "published" .

Publications

pnas

Evaluation of individual and ensemble probabilistic forecasts of COVID-19 mortality in the United States

Cramer, Huang, Wang, Khandelwal · Proceedings of the National Academy of Sciences · · pnas.org

nature

The United States COVID-19 Forecast Hub Dataset

Cramer, Huang, Wang, Khandelwal · Nature Scientific Data · · nature.com

blog

Scaling BeyondCorp with AI-Assisted Access Control Policies

Google Security Blog · · security.googleblog.com

blog

Developing a Distributed Threat Detection Engine

IBM Developer Blog · · developer.ibm.com

patent

System and Method for Value-Based Software Licensing and Copyright Creation using Distributed Ledger Enabled Promotion for Commerce Transactions

Patent № P202200740US01 · google patents

~/background $ cat education.txt leadership.txt

Background

B.S. Computer Science · graduated

University of Massachusetts Amherst

  • Chancellor's Award for highest merit scholarship · Dean's International Award '19 for Excellence in CS
  • Outstanding Leadership Award '19 & '20 · International ACM Student Excellence Award
  • Teaching Assistant: CS326 Web Dev (Spring 2020) & CS397A Predictive Analytics with Python (Fall 2020)
  • Research: software & data engineering for UMass Reich Labs, COVID-19 Forecast Hub with the CDC
  • Industry Deputy: MIT 6.172 Performance Engineering of Software Systems (Fall 2022), invited to teach Fall 2023

Core Organizer · → present

MIT Imagination in Action, with Forbes

  • Secured strategic partnerships with top AI and Web3 firms and curated technical panels
  • Events drawing 4000+ in-person attendees and 100K+ viewers on Forbes & YouTube

Chair, UMass ACM · →

Western Massachusetts' largest student-run tech organization

  • Led the executive board of a 560+ member organization; drove 770% membership growth over two years
  • Sponsorship Committee, MIT Bitcoin Expo: helped raise $200K+ from firms including Binance and Coinbase

~/contact $ cat channels.conf

Contact

Open to conversations about AI security, research, and speaking.

email loading… linkedin linkedin.com/in/ayushkhd github github.com/ayushkhd x/twitter x.com/ayushkhd devpost devpost.com/ayushkhd scholar scholar.google.com/ayushkhd