Skip to content
LiamDGrayPublic

About

Profile

Resources

Stars

1 star

Watchers

0 watching

Forks

Latest commit

 

History

16 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 

Repository files navigation

+-+-+-+-+-+-+-+-+ +-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|E|V|I|D|E|N|C|E| |I|N|F|R|A|S|T|R|U|C|T|U|R|E|
+-+-+-+-+-+-+-+-+ +-+-+-+-+-+-+-+-+-+-+-+-+-+-+

Liam D. Gray

Evidence Infrastructure & High-Assurance AI Systems → liamdgray.com

I engineer deterministic infrastructure for autonomous AI in high-liability environments: extracting fragile signals from the real world, validating them mathematically and cryptographically, and surfacing them as auditable artifacts.

Extract → Validate → Surface

High-Assurance Model Context Protocol (MCP) · High-Performance Rust · Hardened Safety Invariants · Forensic Audit Ledgers · Resilient Ingestion


🏛️ Featured Architectures & Core Systems

  • medplum-mcp — High-Assurance Model Context Protocol (MCP 2.3+) Server for HL7 FHIR Healthcare

    • Dual-stack Python reference server and high-throughput Rust Cargo workspace with Linux pipe zero-copy splice(2) IPC and Axum SSE/HTTP streaming.
    • Compile-time affine typestates and runtime non-bypassable safety gates restricting autonomous AI actions strictly to draft states (zero unauthorized commitments).
    • Deterministic safety gates with Unicode NFKC homoglyph evasion defense and automated FSM reachability invariant verification.
    • 3-tier SIMD token distillation (>89–95% context reduction) and HMAC-SHA256 unbroken hash-chained audit flight recorder.
  • Verus — Hallucination-Free Exhaustive Document Audit Engine

    • Eliminates probabilistic embedding-similarity RAG in favor of exhaustive λ-RLM paragraph analysis for high-liability legal, clinical, and financial diligence.
    • Every claim and conclusion is structurally bound to its primary textual source; hallucination is rendered an architectural impossibility.
  • ScraperForge — Autonomous Scraper Synthesis Compiler & Resilient Runtime

    • Intelligent extraction forge that reverse-engineers target web interfaces, deduces extraction strategies (API hooks, hydration states, DOM traversal), and generates declarative recipe contracts.
    • Integrates semantic dry-run verification and diff-guided LLM self-healing to repair brittle selectors and compile hardened, production-ready collectors.

💼 High-Value Practice Areas

  1. High-Assurance Model Context Protocol (MCP) Infrastructure:
    • Custom dual-stack (Python / High-Performance Rust) MCP 2.3+ servers connecting frontier models (Claude Desktop, Claude Code, Cursor, Windsurf) to regulated systems of record (EHRs, ERPs, clinical registries, financial ledgers).
    • Deterministic safety barriers preventing unintended side-effects and unauthorized state mutations.
  2. Exhaustive Document Audits & Forensic Knowledge Systems:
    • Source-grounded diligence, dose-and-effect mappings, and regulatory compliance verification where missed clauses create legal or operational exposure.
  3. High-Concurrency Systems & Embedded Telemetry:
    • Low-latency systems in Rust, C++, and high-throughput Linux pipe/socket IPC; asynchronous pipeline architecture and cryptographic auditability.

🎓 Background & Lineage

  • Carnegie Mellon University: B.S. Electrical & Computer Engineering (ECE).
  • Qualcomm (#2708): Mobile Station Software (DMSS) — high-reliability embedded wireless communications.
  • Wabtec: Locomotive cab electronic instrumentation (fuel gauges, temperature monitoring, cab radio systems, and telemetry).
  • Core Philosophy: Systems engineering applied to verifiable truth. When correctness is non-negotiable, deterministic engineering beats heuristic vibes.

Website • LinkedIn • GitHub

About

Profile

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors