About

Hi, I’m Leonidas Raghav (most people call me Leo). Welcome to my corner of the internet, where I share insights on AI, software engineering, and the projects I build.

What I Do

I’m an ML Engineer at SAP Labs, where I work on AI agents, infrastructure and developer tooling.

One of my main projects is A3 — a Kubernetes-native platform for orchestrating sandboxed AI agent fleets in enterprise environments. Think infrastructure-first AI safety: assume agents are worst-case malicious or stupid, and bound blast radius through platform design rather than trust.

I also built an internal CLI-based developer tool for coding assistant context engineering that has been adopted by several teams and received a patent filing.

Outside of work, I’ve been involved with AI safety research. I’m a SPAR Spring 2026 Research Fellow, conducting research under supervision on latent sleeper attacks via persistent memory. I’ve also co-authored work that placed 4th at the Apart Research AI Manipulation Hackathon on agent attacks via memory injection, which was selected for the AIMII Workshop at IASEAI’26.

Background

I graduated from the National University of Singapore (NUS) with a degree in Computer Science, where I was part of the University Scholars Programme and made the Dean’s List. My academic and internship experience spanned systems programming, data and AI engineering.

Experience

• SAP Labs Singapore — Building A3, a Kubernetes-native AI agent orchestration platform, and internal developer tooling for coding assistants
• Voltade — Prototyped AI chatbot features and semantic chunking for CRM use cases
• Hitachi Rail — Built streaming ETL pipelines to process real-time train sensor data
• NUS Research — Developed a concurrent web crawler in Rust with a 20× performance gain

Interests

I’m particularly excited about:

• AI Safety & Agent Security — how agent infrastructure, memory systems, and sandboxing interact with safety
• Rust & Systems Programming — safe concurrency, performance, and low-level details
• Cloud Native Infrastructure — Kubernetes, distributed systems, and platform engineering

Projects

Some projects I’ve worked on include:

• A3: Kubernetes for AI Agents — Platform for orchestrating sandboxed AI agent fleets with infrastructure-first safety
• Agent Attacks via Memory Injection — 4th place at Apart Research AI Manipulation Hackathon, selected for AIMII Workshop at IASEAI’26
• MilgramBench — A benchmark for evaluating LLM ethical behavior under institutional pressure
• PaySync — AI financial assistant for SMEs (Text2SQL on invoices)
• Argo — Local-first application for chatting with LLMs offline, built with Rust and React
• RustScript — A statically typed programming language with a bytecode VM

Why This Blog

I started this blog to learn in public — sharing deep dives, engineering lessons, and experiments in AI safety, systems, and software engineering. Hopefully you’ll find something useful here.