1 in 3
PDFs don't have one meaning
502 of 1,572 PDFs produced materially different results across parsers; 43 of 44 IRS tax forms exhibit semantic drift.
How we measured it →35+ years in production · 8 years leading middleware & observability at RGA · now building PQ Crypta & PQ PDF
Senior Systems Engineer & Architect with 35+ years building and operating large-scale infrastructure across insurance, finance and healthcare.
I investigate where machines disagree about reality — from PDFs that mean different things to different parsers to QUIC connections browsers barely explain.
Then I trace each failure to the layer where truth diverges from assumption — and build the systems that expose, measure and explain it.
I'm Allan Riddel. PQ Crypta, PQ PDF and stlweb.dev are mine end-to-end: the Rust, the servers they run on, the monitoring that catches the 2 a.m. failures, and the front ends people actually use. Fated LLC is the company behind them.
Search engines, AI pipelines, compliance systems, browsers and security products all assume machines agree on reality. Often they don't.
In documents, protocols and AI pipelines — I find where systems quietly diverge from what's actually true, then build the tooling that exposes and fixes it.
PDFs that parse differently to every reader silently poison retrieval and training data. I detect the divergence before it reaches the model.
QUIC, HTTP/3 and WebTransport deployments that are misconfigured or silently degraded — found, graded and explained.
Value-vs-appearance drift, parser disagreement and post-signature tampering that break compliance and e-discovery.
Bot and attack traffic that slips past signature-based defenses — caught with WAF rules, JA3/JA4 fingerprinting and ML classification.
Systems that break with no obvious cause. I trace the fault through every layer — app, DB, middleware, network — to the real root.
Moving real TLS and data-at-rest onto NIST PQC (ML-KEM, ML-DSA, SLH-DSA) without breaking what already works.
Across documents, protocols and AI pipelines, the same pattern appears: what users see and what machines process are often not the same thing. I've measured that gap across 24,824 PDFs and live transport analysis.
502 of 1,572 PDFs produced materially different results across parsers; 43 of 44 IRS tax forms exhibit semantic drift.
How we measured it →Of the 16,971-PDF DOJ Epstein release, 18.6% differed between rendered content and machine extraction. AI and RAG pipelines typically ingest extracted text; humans read the rendered page.
Explore semantic drift →Of sites whose servers support HTTP/3, roughly one in four are misconfigured — browsers can't discover it and silently fall back to HTTP/2. The transport tells a different story than the UI.
Analyze your connection →Not slide decks — live platforms, running in production on infrastructure I set up and maintain.
Each platform below rides the same path I designed, deployed and operate end-to-end — a Rust HTTP/3 reverse proxy terminating hybrid post-quantum TLS, carrying QUIC and WebTransport from the browser to the backend.
Infrastructure for the post-TCP web
Rust backend, Python ML pipelines and browser applications. Designed, deployed and operated end-to-end.
Open-source infrastructure
Production Rust infrastructure, publicly auditable, on crates.io / PyPI / npm.
The same PDF can mean different things to different parsers
The 47-engine forensic scanner grew out of that finding — it measures parser disagreement, semantic drift and document integrity across real-world PDFs.
Observability for protocols browsers barely expose
Browsers show pages; transports tell the real story. These tools surface what's actually happening over HTTP/3, QUIC and WebTransport.
Infrastructure-first web platform
Full-stack sites for St. Louis businesses, deployed with HTTP/3, QUIC, hybrid PQC TLS and real observability.
When identical bytes mean different things
Original research showing the document format AI pipelines trust does not always have a single machine interpretation — the same file reads differently to different parsers, measured across 24,824 real PDFs.
I write the backend, run the servers it lives on, trace why it broke, and build the monitoring that catches it next time.
Rust, Python, JavaScript. REST APIs, async services, distributed systems.
Linux & Windows Server, Apache, IIS, F5 BIG-IP, networking, high availability.
TLS, WAF, cryptography, threat detection, JA3/JA4 fingerprinting.
Datadog, custom telemetry, SLOs, anomaly-detection pipelines.
Deployment automation, Jenkins, Buildmaster, reliability engineering.
Tracing failures through every layer until I find the actual problem.
Senior escalation and infrastructure engineering across insurance, finance, and healthcare.
Designed and built the platforms above end-to-end — Rust HTTP/3 proxy and APIs, post-quantum cryptography, a 47-engine PDF forensics scanner, Python ML pipelines, and the infrastructure they run on.
Senior escalation engineer and middleware architect for mission-critical infrastructure across 7 countries. Architected global observability on Datadog, built CI/CD with Jenkins, and ran high-availability infrastructure across IIS and F5 BIG-IP.
Engineering lead for enterprise middleware and application-delivery infrastructure supporting global insurance systems — the same scope continued directly at RGA after the transition.
High-availability web and application infrastructure — IIS, Tomcat, SQL Server, Citrix NetScaler — for mission-critical financial platforms, with HA/DR and disaster-recovery design.
Enterprise application and database infrastructure — WebSphere, SQL Server, SharePoint, ERP — with HA clustering, VMware virtualization, and enterprise backup/recovery.
Systems, network, and IT leadership: data-center operations, multi-org IPSec VPN architecture, NOC management, and custom back-office development for a 10,000+ subscriber ISP.
Senior / staff individual-contributor roles — full-time or contract, remote preferred.
Every path begins the same way — talking to the engineer who built and measured all of this. Where it goes is up to you.
Senior / staff individual-contributor roles in backend, systems & infrastructure, DevOps, or security. Full-time or contract. Remote preferred.
Bring me in for QUIC / HTTP/3 / WebTransport, post-quantum TLS, or proxy and observability work — or license the engines directly.
Start by talking to the engineer who measured semantic nondeterminism across 24,824 PDFs and built observability for the post-TCP web. If it goes further, the platforms, source and brands — PQ PDF, PQ Crypta, the QUIC/WebTransport suite — are available, together or apart.
Yes. I'm open to senior/staff individual-contributor roles — full-time or contract, remote preferred — and can start quickly.
Systems & infrastructure, backend (Rust), security, protocol/transport (QUIC, HTTP/3, WebTransport), observability/SRE, and research engineering.
Yes — remote is preferred, with openness to hybrid for the right role.
Yes — full-time, contract, advisory, or licensing of the engines (the Rust proxy, post-quantum cryptography, and PDF forensics).
Yes — the platforms, source, and brands are available to the right buyer, together or apart, including acqui-hire of the person who built them.
If you're exploring QUIC / WebTransport, AI-ingestion safety, document integrity, protocol observability, PDF forensics, post-quantum TLS — or acquiring PQ PDF / PQ Crypta — let's talk. One email reaches the engineer directly: no recruiter layer, no funnel.