I build the infrastructure AI coding agents need to work on real codebases — structured code understanding, persistent memory, context economy, and architectural constraint. Twenty years in enterprise infrastructure; currently a validation engineer working on VMware and Linux network driver certification.
| Project | What it does |
|---|---|
| CodeGraph | Semantic graph of your codebase — functions, classes, imports, call chains — exposed as 42 MCP tools across 38 languages. One Rust binary serving MCP and LSP; VS Code extension, JetBrains plugin, npm package, GitHub Action. |
| Tempera | Episodic memory for coding agents. Captures sessions as episodes, retrieves them by semantic search, and uses reinforcement learning to surface what actually helped. |
| Smelt | Semantic version control over Git. Record the intent, get a semantic delta — functions changed, breaking signatures, dependency impact — validated against architectural rules before commit. |
| Crucible | Architecture validation for AI-generated code. Checks implementations against a machine-readable system definition to catch drift before it lands. |
| Compressor | Token reduction for Copilot agent mode. Compressed reads, searches, and outlines as Language Model Tools, with recoverable expansion markers and a local savings ledger. |
All Rust unless noted. Apache-2.0.
- CVE-2026-50163 (High) — link-following path traversal in
oras-gotar extraction. CNCF ORAS project; fixed in 2.6.2. - CVE-2026-47699 — escaping hardlink during OCI image unpacking in Confidential Containers
image_rs; fixed in 0.20.0. - Microsoft Security Response Center — bounty awarded, researcher recognition Q2 2026.
English · Russian · Hebrew



