I build low-level systems, local inference/ML tooling, and the infrastructure that lets AI agents do real work. By day I work on high-performance in-memory data systems at AWS ElastiCache; the OSS below is my own.

• agent-workspace-linux - isolated, agent-owned Linux desktops over MCP: a hidden desktop and browser an agent can drive for GUI and web work without ever touching your real machine.
• computer-use-linux - Linux desktop control over MCP via AT-SPI, GNOME Shell, Wayland portals, and ydotool. Shipped through Codex Desktop Linux, including the performance and read-aloud work.
• agnix - the missing linter + LSP for AI coding-assistant configs. Rust core with a GitHub Action, MCP server, and editor plugins; 432 rules over CLAUDE.md, AGENTS.md, SKILL.md, hooks, MCP, and plugin metadata, with autofixes.
• agent-sh / agentsys - the org I run: a DX-focused ecosystem for AI-assisted development. 24 plugins, 49 agents, 44 skills, 24 commands, phase-gated workflows; works with Claude Code, OpenCode, Codex, Cursor, and Kiro.
• eigen - local agent harness in Go with both a GUI and a CLI: one daemon, many resumable sessions, multi-provider routing (Bedrock/Claude/OpenAI-compatible/local), memory, and observability.

• Valkey GLIDE - maintainer of the official multi-language Valkey/Redis client (Rust core, Java/JNI, Node/N-API): reliability fixes, TLS, timeout/retry behavior, API review, CI, releases, and community.
• Valkey - contributor to the in-memory datastore core; started the official valkey-skills.
• rustowl - performance and reliability work on the ownership/lifetime visualizer (5k★ Rust): runtime/stack-size refactor, jemalloc integration, memory fixes, benchmarking, and CI/security-testing workflows.
• FlowFabric - Rust durable-execution engine for Valkey/Postgres/SQLite: lease-safe workers, waitpoints, HMAC-signed human approval, capability routing, budgets, and streaming output.
• Low-level tooling - layout-audit (DWARF binary memory-layout analysis: padding, layout diffs, size budgets for C/C++/Rust/Go) and scrump (format-aware secret scrubber for binary capture artifacts).

Local-LLM serving and inference performance: speculative decoding / MTP, MoE expert routing and surgery, KV-cache quantization, and long-context serving on consumer GPUs.

Beyond GLIDE and core, I build infrastructure around Valkey: glide-mq - a high-performance Node.js queue on Valkey/Redis Streams (Rust N-API core, AI-native primitives) with a full plugin family: framework adapters for Hono, Fastify, Hapi, and NestJS, plus a dashboard and examples. Also ocaml-valkey (OCaml 5 + Eio client on opam, RESP3-only, cluster routing with AZ-aware reads, TLS/mTLS, Search/JSON/Bloom wrappers).

I spend real time on the support side of OSS: issue triage, reproductions, user debugging, CI/release health, and helping contributors land better PRs. On Stack Overflow: 1,130+ reputation, strongest tags redis, github-actions, node.js, amazon-web-services, valkey.

• Inside Valkey GLIDE — AWS Developers Podcast
• Glide into resiliency — Let's Talk About Data
• How AI workloads changed the queue I was already building · Your AI Agent Configs Are Probably Broken

aviarchi1994@gmail.com — systems, datastore internals, inference/ML, Valkey/Redis clients, agentic tooling.

If this work saved you time, sponsorship helps me keep doing it.