mlc-llmAMD · gfx1100
Qwen3-8B-abliterated-q4f32_1-MLC
106.0tok/s
8 GiB VRAM
Integration architecture and reliability systems
Cody Blevins
I build healthcare integration systems: HL7v2, FHIR, and X12 pipelines, the validation gates in front of them, and the observability that catches silent failures before a partner does.
Integrations Manager at Devoted HealthHealthcare interoperabilityPlatform reliability
Current role
Devoted Health
Integrations Manager
Healthcare integration delivery: EDI and API exchange with partner vendors, monitoring, and issue triage.
Start Here
Start with what you need
Pick the shortest path to what you came for.
Selected Work
fi-fhir, edilint, FlexInfer, Loom, and the platform they run on.
Resume
Full delivery history, with lenses for different role tracks.
Writing
Implementation notes and case studies.
About
Career history and operating principles.
Playground
Live demos: parse messages, build mappings, run tooling.
Products
Docs and product pages for the tools I maintain.
Services
Consulting engagement shapes and scope.
Contact
Email works best. I read everything.
Professional Focus
Healthcare integrations, built and operated
I design, deliver, and operate healthcare data exchange, and I build the tooling that keeps it reliable.
14Years in Healthcare IT
Healthcare Data Exchange
HL7v2, FHIR, X12 EDI, and API integrations across athenahealth, Clover Health, and Devoted Health.
From interface-level support to running API support engineering for an integration network
2Open-Source Healthcare Tools
Integration Tooling
fi-fhir parses legacy formats into semantic events; edilint gates outbound EDI and flat files in CI.
Both in Go: static binaries, tested, documented, inspectable
K8sProduction Homelab
Platform Engineering
K3s on Harvester, Flux GitOps, Prometheus/Grafana/Loki, self-hosted GitLab CI, and GPU inference.
Everything on this site builds and ships through it
Proof of worksnapshot
Measured on the homelab runtime
FlexInfer is the private inference platform I keep running on real hardware. These results come from the cluster benchmark records, not a sample dataset.
Snapshot verified
106.0tok/s recorded peak
mlc-llmAMD · gfx1100
Qwen3-14B-q4f16_1-MLC
82.4tok/s
16 GiB VRAM
mlc-llmAMD · gfx1100
Qwen3-14B-abliterated-q4f16_1-MLC
75.0tok/s
16 GiB VRAM
teiCPU
BAAI/bge-large-en-v1.5
203.3emb/s
2 GiB memory
diffusersAMD · gfx1100
stabilityai/sdxl-turbo
61.6img/min
10 GiB VRAM
Snapshot scope. These legacy benchmark-result records preserve backend, hardware class, memory, and throughput, but not prompt, batch, or sample counts. Treat them as operational baselines, not a cross-vendor leaderboard.
Inspect the source recordsSelected Work
Projects & Labs
Live projects with demos and source. The full list is on the selected-work page.
Writing
Interoperability, Reliability, and Delivery
Practical notes from healthcare integrations, platform reliability work, and cross-functional delivery.
August 29, 2026
The Agent Gets Read-Only Credentials: A Trust Ladder for AI in Regulated Systems
How I run LLM agents against regulated healthcare payment data without losing sleep: a three-tier trust ladder, kill-tests before promotion, and guardrails that live in infrastructure instead of prompts.
ai-agentsllmguardrailsmcphealthcaretooling
August 14, 2026
The File That Failed With Zero Errors: A Field Guide to Invisible EDI Defects
Four defect classes that healthcare interchange parsers accept silently, with synthetic samples, real linter output, and the pre-send gate that catches them.
healthcareedix12validationtoolingopen-source
Evaluating my work?
The selected-work page links to source, the resume covers the history, and email is the fastest way to reach me.