AI Systems Engineer · LLM Agent Architect. I build production systems around LLMs — RAG services, agent orchestration, semantic search, self-hosted inference — under real latency and cost constraints, and I run them in production.

🔭 Building Protocore — a protocol-first agent runtime for small, self-hosted LLMs. The loop, the context budget, the tool surface, the compaction and the stop conditions; direct and deep run strategies, delegation with real width and depth budgets, snapshot and resume. A library, not a platform — you embed it in your own API. The platform on top of it (enterprise backend, sandbox, chat, dashboard, dynamic tools over MCP, proto terminal client) is proprietary and runs on my own GitLab / Harbor / k3s.

📖 The core is open source under MPL-2.0 → anchor-inference/protocore · pip install protocore==2.0.0a25. Zero upward imports, no database driver, no HTTP endpoint — everything outward-facing is a Protocol you implement. 3 200+ tests, 90% branch coverage enforced, strict typing, gated on Python 3.12 / 3.13 / 3.14.

🤖 Daedalus — a personal self-developing agent on that core: its own app window on macOS, Linux and Windows, a web app on your phone and Telegram if you want it; real tools, projects run by an orchestrated team of subagents, and changes to its own code as pull requests you approve. MIT · daedalus.anchorinference.com · 3-minute film: English / Russian.

🧰 Smaller tools for agent workflows: ai-audit-kit · kb-genesis · notify-telegram-cli · swiftclf-tuna-research

🛠 Stack: Python · FastAPI · Pydantic v2 · asyncio · vLLM · OpenAI-compatible APIs · Elasticsearch / OpenSearch · Postgres · Redis · RabbitMQ · Docker · Kubernetes / Helm · gVisor · Next.js / React

📍 Astana, Kazakhstan

💬 Telegram: @notsoulmate

📫 Email: ascorblack@gmail.com · a@scorblack.ru

🌐 Site & resume: ascorblack.com · ascorblack.com/resume