Hi, I'm Aayush Kumar
|
I build reliable AI infrastructure at scale — from production RAG systems and agentic security copilots to research on multimodal safety and structured outputs.
What I Build
Shipping AI systems that work in production — not just demos
AI Security
Building agentic security systems that find and fix vulnerabilities automatically
- <3% false positives
- 30%+ auto-merged fixes
- 50+ PRs/week
RAG & Agents
Production pipelines that are reliable, fast, and hallucination-resistant
- 18% fewer hallucinations
- Schema validation
- Streaming parsing
ML Systems
High-performance infrastructure that scales without breaking the bank
- 6× compute reduction
- 220GB → 52GB memory
- 99.9% uptime
Impact at Scale
Featured Projects
Production systems, research prototypes, and open-source tools
HyperSentry
An agentic security copilot that uses policy checks + retrieval to generate remediation PRs and accelerate incident analysis.
- False positives under 3%
- Auto-merged 30%+ remediation PRs
SchemaPulse
OpenAI-compatible inference layer for schema-valid structured outputs with streaming tool-call parsing.
- Streaming incremental tool-call parsing
- Adaptive guided decoding for schema validity
EdgeJury
Multi-LLM council with cross-review for truthful QA on serverless edge inference.
- Cross-reviewed small-model ensembles
- Structured critique + synthesis
Ghosts in the Scale
Prompt injection via image resampling & certified defenses for multimodal systems.
- Black-box downscaler fingerprinting
- ScaleJail-mini benchmark
Publications & Patents
Peer-reviewed papers, preprints, and awarded patents
EdgeJury: Cross-Reviewed Small-Model Ensembles for Truthful Question Answering on Serverless Edge Inference
arXiv
SkinGuardian: On-Device AI for Private, Fair, Robust, and Explainable Skin Cancer Detection
Research Square
HybridNet: Advancing Deepfake Detection Through Residual, SE, and Depthwise Convolutions
IEEE Access, Vol. 13
What I'm Building
Focused on making LLM systems reliable, secure, and efficient. Current themes include structured output guarantees, multimodal safety, and intelligent budget allocation for reasoning.
AI Security
Adversarial robustness, prompt injection defenses, and secure agentic systems
Reliable Inference
Schema-valid outputs, streaming parsing, and structured tool calling
Let's build something together
Seeking ML Engineer roles in the USA, open to global remote.