Rick Huang · Applied AI · Data
Better data models. AI that does real work.
I'm Rick Huang, an Applied AI Architect, Business Lead. I design the data models businesses run on and put AI to work inside them. Here I share what holds up in practice.

- RoleApplied AI Architect · Data Architect · Business Lead
- AdvisorIndependent Technology Consulting
- WritingTechnology Investment and Thoughts on Life
- IndustriesConsumer Product · Manufacturing · Healthcare · Financial Services · SaaS
Start here
Four essays that explain what I do.
Data ArchitectureOne Revenue Question, Three Answers: Why AI Needs a Semantic LayerARR, bookings or revenue? One SaaS deal has three correct answers. Why your AI reports the wrong one, and how a semantic layer fixes it for good.Applied AIWhy Your AI Video Budget Isn’t Turning Into AdsAI tools made video drafts cheap. Ads your brand will approve are as hard as ever, and every rejected try is billed again, often at a markup on the model.Data ArchitectureBefore You Build the Agent: Working Backwards to Your Data PipelineBefore an AI agent can answer for the whole business, every sales channel's numbers have to agree. Part 1: how one company stopped stitching spreadsheets.Living LifeEveryone has AI now. So why would anyone pick yours?When one feature has a hundred look-alike products, the product stops being the reason people choose. A person they trust is.
For data teams
Four essays on how it's built.
Applied AIThe semantic layer is the API for AIAI that writes its own database queries looks great in demos and quietly gets finance numbers wrong. Give it approved metric definitions instead.Applied AIEvals for Analytics Agents: Grading Answers Against the LedgerAn analytics agent that is 85% accurate is useless in finance until you know which 15% is wrong. Build a golden set and gate every change.Applied AIData Models Agents Can TrustAnalysts fill gaps in a data model from experience; agents fill them with confident guesses. Four fixes, with real Salesforce examples.Applied AICertifying Data Models with an Agent in the LoopNever let an AI agent decide whether finance data is fit to use. Rules decide, a safe test runs on real data, and the agent only explains failures.
Latest
New on the Blog
- I Gave the Router the Big BrainDeepSeek keeps its router thin. I gave mine the strongest model, added an advisor, and let cheaper workers build. One real run, with the token numbers.
- Microsoft Is Offering Its Own Copilot Team a Buyout. And That's Your AI Strategy?A company about to lose 12% of its revenue banned every AI tool except Copilot. Meanwhile, Microsoft is offering its Copilot team a check to leave.
- Two Dashboards, One Number, Zero TrustSales says $12.4M, Finance says $9.8M. Why two dashboards disagree, and four plain-language fixes that make company numbers trustworthy again.
- Your AI Shouldn’t Need a BabysitterDeepSeek’s Super-agent design says the limit on AI isn’t the model anymore. It’s my attention. How it works, and how I already run my week this way.
- Less Is More: Cut the Data Catalog, Keep Its JobData catalogs like Atlan and Alation were built so people could find and trust data. AI changed who is asking. Cut the copy, keep the job it did.
Finance data
Five essays on making the numbers agree.
- Ledger-First: A Data Architecture for a Crypto CustodianOps, finance and growth report three different AUM numbers. Derive every one from a single double-entry ledger, and reconciliation becomes a test.
- One Customer, Five Systems: Why Your Top-Client Report Is OffCRM, contracts, ERP, billing and KYC each count customers differently. Resolve them to one party without guessing, and prove it ties out.
- Basis Points on a Moving Balance: Modeling Asset-Based FeesSame client, same balances, same rate card: four methods bill from $2,740 to $3,879. How to model basis-point fees so the invoice is right.
- As-Of vs. As-Was: Why December 31 Has Two Right BalancesLate prices and backdated deposits give December 31 two right balances. How to model both in dbt, and why auditors care which one you show.
- Migrating a Chart of Accounts Without Breaking a Single ReportRe-cutting the general ledger exposes every model, metric and dashboard downstream. Map it as data, dual-run it, reconcile every node to zero.
Need a speaker on applied AI and the data underneath it?
Conference talks, company offsites, podcasts and private workshops. Practitioner-level, first-person, no slides full of buzzwords.