I build security infrastructure for AI systems and explore the intersection of technology with classical Indian knowledge traditions.
Currently: Building Highflame — an AI security platform that protects LLMs and agents against adversarial threats with sub-10ms guardrails, Cedar-based policy enforcement, and taxonomy-driven vulnerability detection.
Previously: Data engineering, distributed systems, and large-scale ML pipelines.
AI Safety & Security — Prompt injection detection, jailbreak defense, agentic safety (loop detection, tool call validation, MCP security), credential leakage prevention. Fine-tuned 16+ transformer models for threat classification.
Vedic Computing — vibz.art — a creative platform powered by Lilavati, a Rust-based panchang computation engine that models astronomical cycles with mathematical precision. Bridging ancient Indian calendar systems with modern software.
Open Source — I believe in building in the open. My work spans guardrail engines, policy frameworks, ML model pipelines, and content generation systems.
Languages Rust · Go · Python · Scala · TypeScript · C++ · Java
ML/AI PyTorch · Hugging Face · LangChain · Fine-tuning · RAG
Data Spark · Kafka · Flink · ClickHouse
Infra Kubernetes · Docker · Prometheus · OTEL · Cedar
I write about things I'm learning — AI safety, systems architecture, LLM research, and occasionally philosophy.
yd2590@columbia.edu · Always happy to talk about AI security, distributed systems, or Sanskrit computational linguistics.