Skip to content
View r-irbe's full-sized avatar
🎯
Focusing
🎯
Focusing

Highlights

  • Pro

Block or report r-irbe

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
r-irbe/README.md

Radu Irbe

Staff engineer and researcher. Distributed systems, AI infrastructure, developer tooling, formal verification.

I like making implicit assumptions visible and testable.

Public

  • proof-skills — Lean 4 / Mathlib4 agent skills, eval suite, Glicko-2 comparison
  • lean4-lsp-mcp — MCP server for live Lean 4 proof state and C FFI
  • lifecycle-guard — Fail-closed policy engine for agent tool calls (allow / warn / escalate / block) (currently spec)
  • queue-reactive-book — C++20 limit-order-book and queue-reactive market-making simulator
  • hexcrawler — TypeScript crawler, hexagonal architecture, SSRF-safe redirects
  • r-irbe/ik_llama.cpp — Contributions to high-performance local LLM inference, focusing on SOTA low-bit quantizations and CPU/GPU hybrid execution.
  • r-irbe/pi-lens — Contributed Lean 4 language support for the LSP-powered extension giving AI coding agents real-time, language-aware feedback to ground reasoning and reduce hallucinations.

Projects & Research

Much of my recent work lives in private repositories or enterprise environments, spanning planetary-scale backend systems, rigorous AI governance, and formal methods:

  • Strictly Governed AI Workspaces: Architected self-evolving, spec-driven agent harnesses for deterministic human-AI collaboration. Built private swarm governance (pi-fleet), fail-closed lifecycle safety nets (pi-harness-bridge), and traceable peer-review engines (pi-review-kit).
  • Planetary-Scale Distributed Systems: Architected and designed the platform observability, load management, QoS for a new CQRS based data platform, targeting >1B reads/hour with <2ms latency at >99.99% uptime (Microsoft). Managed Kafka/Flink stream processing pipelines executing zero-downtime, terabyte-scale 5-way joins on shared Tier-0 infrastructure (Wise).
  • Formal Verification of High-Risk AI: Formally verifying AI system security properties using Lean 4 for high-risk applications.
  • High-Performance Engineering: Built deterministic platforms including a memory-mapped NVMe-resident graph engine (rust-mmap-engine) constrained by a Lean 4 rule engine, and mathematically verified decision engines.

Looking for work that mixes infrastructure with controlled AI-assisted development.

irbe.ai · LinkedIn

Pinned Loading

  1. proof-skills proof-skills Public

    APM-installable agent skills for Lean 4 and Mathlib4 — proof tactics, math domains, review and research workflows, generic tooling.

    Python 2

  2. lean4-lsp-mcp lean4-lsp-mcp Public

    Standalone Lean 4 + C FFI Model Context Protocol (MCP) server

    TypeScript

  3. hexcrawler hexcrawler Public

    Horizontally scaled TypeScript crawler: hexagonal architecture, SSRF-safe redirects, per-domain circuit breakers, 1100+ tests.

    TypeScript

  4. lifecycle-guard lifecycle-guard Public

    Fail-closed policy engine for agent tool calls: allow / warn / escalate / block, hash-chained audit, session resume from protobuf and SQLite.