Software engineer working at the intersection of AI and developer tooling — I build open-source libraries and tools around LLM safety, security automation, and context engineering for AI agents.

Most of my current open-source work is about making LLM applications safer and smarter:

• injectbench — prompt-injection regression testing for LLM apps: a curated attack corpus plus deterministic graders, zero-dependency and drops straight into CI
• llm-cloak — a context firewall for LLM apps: redacts secrets and PII before text reaches a model, restores originals on the way back, and strips injection payloads
• open-context-engine — an open toolkit for building context-aware AI agents: AST-level code understanding (tree-sitter), retrieval, and incident workflows across a VS Code extension, TUI and studio
• security-agents — local-first Solidity security analysis: Slither findings ingested, exploit hypotheses generated and validated without sending code to a cloud model
• ai-skills-library — a curated library of AI skills, prompts, workflows and automation patterns

• super-utils-plus — zero-dependency TypeScript utility library in the spirit of Lodash: 140+ fully-typed functions (npm)
• yt-caption-kit — fetch, translate and format YouTube captions and transcripts: typed API, CLI, SRT/WebVTT output (npm)
• better-mitt — tiny fully-typed EventEmitter for browser and Node (npm)
• lotame-sdk — TypeScript SDK for Lotame Lightning Tag with browser and SSR helpers (npm)
• mynota — minimal browser text editor where the whole note lives deflate-compressed in the URL — shareable, no backend

LLM engineering — prompt-injection red-teaming and defense, context firewalls (secrets/PII redaction), context engineering for coding agents, MCP, evaluating and regression-testing LLM apps in CI AI + security — scam and malware scanning for crypto tooling, static-analysis triage (Slither), AI-assisted exploit-hypothesis workflows Code intelligence — tree-sitter/AST-based chunking and dependency graphs, embeddings and retrieval, fast local-first tooling Languages & platforms — TypeScript, Node.js, Python; browser, CLI and editor tooling

Deep in AI safety tooling and agent infrastructure — better injection defenses, smarter context systems, and security automation that runs locally. I like solving problems people hit every day and shipping the fix as open source.

Open an issue or discussion on any of the repos above — happy to collaborate on LLM safety, AI tooling, or security automation.