Allen Huang.

Data Architecture · AI Engineering

The data foundation your AI actually needs.

I'm Allen. I design and build cloud data platforms — Snowflake, dbt, BigQuery, Fivetran — and then build the AI that runs on top of them. Four warehouses from scratch, two of them for Berlin teams.

Book a free call See my work →

Available for new projects · Porto-based · EU & US time zones

Portrait of Allen Huang

Allen Huang — Porto, Portugal

4 Warehouses Built

Cloud data platforms designed from scratch — three of them on Snowflake — across US healthcare and Berlin scale-ups.

58 Models, ~200 Tests

A layered staging-to-marts dbt architecture, gated by automated data-quality tests, delivered for one Berlin client.

$10s of M Recovered

Revenue surfaced for a US healthcare client through billing analytics and revenue-cycle automation.

MSc, Distinction

Computer Science, University of Cape Town. Published at IEEE CEC & AAMAS on ML and multi-agent systems.

The problem

AI doesn't fail at the model. It fails at the warehouse — wrong grain, metrics that disagree, and numbers the exec team has quietly stopped trusting.

I work both layers. I architect the platform — ingestion, modelling, tests, metric definitions — and I build the AI that sits on top of it. Same person, same system, no handoff between the team that owns the data and the team that owns the model.

What I do

Services

From the warehouse up. Pick the layer you need help with — or hand me the whole stack.

01

Data Platform & Warehouse Architecture

Design and build your cloud warehouse from the ground up — Snowflake or BigQuery, ingestion through to marts. Layered, documented, and tested, so the numbers hold up when someone senior pushes back on them.

  • — Warehouse design & dimensional modelling
  • — ELT & change-data-capture ingestion
  • — Cost, performance & access design

02

Analytics Engineering (dbt)

Turn raw source tables into a modelling layer your team can actually extend — staging through to marts, with data-quality tests as a gate in CI rather than an afterthought nobody runs.

  • — Layered dbt architecture & conventions
  • — Data-quality testing in CI
  • — Metric & semantic definitions

03

Migration & Modernisation

Move off manual spreadsheets, legacy BI, and hand-maintained extracts onto a cloud warehouse — without breaking the reporting the business runs on while you do it.

  • — Legacy BI & Power BI migration
  • — Spreadsheet → warehouse workflows
  • — Parallel-run & cutover planning

04

AI Integration & LLM Engineering

Embed LLMs into your existing product or workflow — conversational interfaces, document processing, automated decision support. Built for production: typed, tested, observable.

  • — Conversational AI & chatbots
  • — Document generation & extraction
  • — Agent workflows & tool use

05

Specialist

Healthcare Data & AI

Revenue-cycle analytics, clinical transcription pipelines, and structured clinical data extraction. Experience with US patient data, encounter records, CCM/RPM billing workflows, and HIPAA-adjacent systems.

  • — Revenue-cycle & billing analytics
  • — Clinical NLP & structured extraction
  • — Patient data modelling over time

06

Fractional Data & AI Lead

Embed me in your team on a retainer — set the data and AI roadmap, evaluate tools and vendors, build internal capability, and ship incrementally. I've built and hired a BI department before; no full-time hire needed to start.

  • — Roadmap & tool selection
  • — Hands-on implementation support
  • — Hiring, enablement & documentation

Selected work

Production systems I've built

Client names withheld for confidentiality — happy to talk specifics on a call.

Warehouse from zero

Berlin Subscription Software

Designed and built the company's first cloud data warehouse from the ground up — production database streamed in by change-data-capture, alongside billing, auth, and product-analytics sources. Delivered 58 models in a layered staging-to-marts architecture guarded by close to 200 automated tests, including a full subscription-revenue model: MRR history, movement split into new, expansion, contraction and churn, and revenue net of credit notes and refunds. Added a containerised FX service so multi-currency reporting reconciles, and per-generation AI cost-of-goods so finance can finally see gross margin per feature.

BigQuery dbt Cloud CDC Subscription Revenue FinOps

Power BI → Snowflake

Berlin Consumer App

Migrated a consumer scale-up off manual Power BI workflows and onto a modern cloud stack — Fivetran ingestion, dbt modelling, Snowflake warehouse. Rebuilt the customer metrics the business actually steers on: segmentation, lifetime value, and retention, defined once in the modelling layer so marketing, product, and finance stop arguing about whose number is right. Also tightened paid-acquisition attribution across the mobile and web ad platforms.

Snowflake dbt Fivetran Migration Marketing Analytics

BI from scratch

US Healthcare Startup

Built the organisation's BI function from scratch: hired the team, implemented the full analytics stack (Snowflake, Fivetran, dbt, Sigma Computing), and automated tracking of 20+ KPIs across 10 departments. Developed insurance billing dashboards surfacing unbilled, underpaid, and denied claims — recovering tens of millions of dollars in revenue. Also delivered near real-time P&L tracking for the finance department.

Snowflake dbt Fivetran Healthcare Revenue Cycle

Clinical AI pipeline

US Healthcare Platform

Designed and built a production AI pipeline for a care-management platform that turns recorded patient calls into structured clinical insight: automated transcription, LLM-based extraction of clinical and social-need signals, and a clinician review step to verify results before they're used. Engineered for real-world reliability — asynchronous processing, graceful failure handling, and end-to-end monitoring.

Healthcare LLM NLP Speech-to-Text Python

AI cost & observability

AI-Native SaaS

Built a token-usage and cost-tracking model for a company running large-scale, multi-provider LLM infrastructure. The hard part was correctness: normalising inconsistent billing data across providers and accounting for prompt-caching properly — fixing the cache accounting alone corrected a roughly 30% overstatement in one provider's costs. Delivered a finance-grade cost model with per-call attribution, versioned pricing, and automated validation tests.

AI Observability dbt BigQuery FinOps

ML at scale

Major Financial Asset Manager

As Technical Lead at one of Africa's largest asset managers, built and led a data science team applying ML to improve business processes. Delivered a clustering-based behavioural segmentation toolkit used by business development managers to personalise advisor relationships, and a neural network-based workflow automation system. Operated at institutional scale across a five-year engagement.

Machine Learning Data Science Financial Services Python

How it works

A simple way to start

  1. 01

    Discovery

    A 30–45 min call to map your data estate, your AI goals, and the constraints around both.

  2. 02

    Proposal

    Scope, milestones, and success metrics. Fixed-price or time & materials.

  3. 03

    Build & Enable

    Ship iteratively, document thoroughly, enable your team to own it.

Tools

Tech stack

Data platform

Snowflake dbt BigQuery Fivetran Datastream (CDC) SQL Python AWS GCP

AI & LLMs

Claude (Anthropic) Gemini GPT AssemblyAI LangChain

BI & engineering

Sigma Computing Looker Metabase Power BI Django Celery ReactJS

About

A builder, not just an advisor

I'm Allen. I build the data platforms companies run on — warehouses, ingestion, modelling layers, and the tests that keep them honest — and increasingly the AI that sits on top of them. Four of those platforms I've built from scratch; three of them run on Snowflake.

My background in AI goes back further than the current hype cycle. I hold an MSc in Computer Science (with distinction) from the University of Cape Town, where my research focused on neuro-evolution for collective autonomous systems. I've published at IEEE CEC and AAMAS, lectured Machine Learning at university level, and spent five years leading a data science team at a major financial institution before going independent.

I've also built and hired a BI department from nothing, and built and exited my own e-commerce business — so I understand both the org chart and what founders are actually dealing with.

I'm based in Porto, work remotely across EU and US time zones, and keep my client list small so I stay hands-on.

Contact

Let's build something.

Tell me about your data estate or your AI problem — usually they're the same problem. I'll reply within one business day.

Email me Book a 30-min intro call →