Applied AI Architect · Building in production

Charles Kim
I build AI that ships.

If your AI project is stuck in pilot purgatory, I'm the person you call. I build infrastructure that runs in production — eight brands, 18 countries, 25 million+ interactions a year — and consult with teams who need it to actually work.

64× Cost advantage, in-house RAG vs. vendor POC
$1M+ Annualized compliance automation savings
25M+ Customer interactions per year
18 Countries in production
RAG pipelines Multi-agent systems Vendor evaluation Compliance automation Genesys Cloud OpenRouter ElevenLabs Claude API Production AI CX Architecture Golden test harness Voice infrastructure SSOT pipelines Agentic workflows RAG pipelines Multi-agent systems Vendor evaluation Compliance automation Genesys Cloud OpenRouter ElevenLabs Claude API Production AI CX Architecture Golden test harness Voice infrastructure SSOT pipelines Agentic workflows
Proof of work
Numbers that are real
64×
Cost advantage
Built the financial model comparing in-house RAG against the Sierra AI vendor POC. Ran the automated test harness. Validated by Finance.
$1M+
Annualized savings
CCPA and GDPR compliance automation, designed and deployed in-house. Savings validated independently by Finance. Zero vendor dependency.
57
Golden test cases
Quality scoring framework for RAG output — three classifications, production-validated, built to catch real failure modes not demo scenarios.
25M+
Interactions / year
Customer interactions running across HelloFresh Group infrastructure. Eight brands, 18 countries, one architecture.
140+
Semantic chunks
Global knowledge base spanning six documents and 55 evaluation questions. Built with a Node.js generation pipeline, deployed to production.
18
Countries
HelloFresh Group footprint: HelloFresh, Factor, EveryPlate, Green Chef, Chefs Plate, YouFoodz, Good Chop, The Pets Table.

Hover each card for context.

Side projects
Things I build outside work
autofiller.app AI

Job application copilot. Voice-matched cover letters and ATS field capture. Each application gets a tailored pass, not a bulk blast.

PHP · Node · Playwright · Claude API
megadash.io Tool

Privacy-first browser dashboard for AI-generated artifacts. Runs entirely in the browser — no server, no tracking, no login.

React · Babel · localStorage
repairlayer.com Ops

Repair shop operations platform with three roles: intake, technician, and customer status. Built in one day, running in the field.

PHP · MySQL · PWA
adminlayers.com Infra

Tooling layer for Genesys Cloud administration and Google Workspace automation, with AI orchestration built in.

Node · Genesys · Google APIs
ai4noco.com Consulting

Applied AI consulting practice for Northern Colorado. Workshops, literacy sessions, and hands-on production builds.

Strategy · Education · Builds
nexus-orchestrator AI

Multi-agent platform where one agent challenges, one defends, and a third synthesizes the output. Evaluated across 48 test categories.

React 19 · TypeScript · Vite · OpenRouter
relatedmodels.com Blog

Model recommendation service and editorial. Find the right AI model for your use case — reviews, comparisons, and picks from someone who runs them in production.

PHP · OpenRouter · Editorial
modelmix.app AI

Multi-model routing and comparison. Pick the right model for the job without changing your workflow.

PHP · OpenRouter
megatts.com Voice

Text-to-speech utility infrastructure. Fast, clean, deployable to any shared host.

PHP · TTS APIs
Featured work
Multi-agent debate architecture
Challenger
COG CORE
Attacks assumptions, finds failure modes, and stress-tests every position before it can be accepted.
Defender
COG PRIME
Builds the strongest possible case for the position under scrutiny. Holds the line under pressure.
Synthesizer
Orchestrator
Evaluates the debate between the other two and produces a grounded, verified output. No agent wins alone.
Output scored across 48 test categories — verified in production at HelloFresh Group

Single-model systems produce confident answers even when wrong. Running agents in structured disagreement surfaces those gaps before they reach customers. The logic is simple: if the challenger can't break the defender's position, the output is trustworthy. If it can, the synthesizer catches it.

The architecture is model-agnostic by design. Models are swappable. The debate structure and evaluation layer are the actual asset.

Experience
Career

I've never worked somewhere that handed me a finished system and asked me to maintain it. Every role has been some version of: here's a mess, figure it out. That's not a complaint — it's where I learned everything. At Dell it was a phone system sprawl that had outgrown anyone's ability to understand it. At KPMG it was a federal healthcare operation standing up in real time during COVID. At HelloFresh it started as platform ownership and became something I didn't have a job title for yet: the person who decides whether the AI vendor is telling the truth, builds the alternative if they're not, and ships it before the next budget cycle.

2019 – present
Senior Manager, CX Technology & Automation
HelloFresh Group
Applied AI architect across eight brands. Built RAG infrastructure, automated compliance workflows, evaluated and benchmarked AI vendors, designed multi-agent systems, and own the full CX technology stack across 18 countries.
2020 – 2021
Manager, Federal Healthcare BPO
KPMG
Manager-level federal healthcare contact center operations during COVID. Three direct reports. Large-scale CX operations under real pressure with real stakes.
Prior
Enterprise CX Architecture
Dell / NTT DATA
Dell: consolidated 140+ PBX systems into 7 virtual ACDs. ICM and WFM lead. NTT DATA: enterprise contact center consulting and architecture across multiple verticals.
2002 – present
Founder
Web Hosting Reseller
Running a reseller hosting business since the early web. Deep WordPress and LAMP stack experience. The first side project.
Let's talk

Based in Berthoud, Colorado. Remote-friendly.