The whole company asks the data questions.I run the platform that answers.

I'm a data & AI platform engineer at a specialty insurer. I own the Kafka CDC streaming backbone, the governed Snowflake lakehouse, and the enterprise Claude plugin marketplace I spearheaded — 12 plugins, authored domain skills, and model evaluations across six business domains, used analyst to C-suite to answer business questions in plain English. Agentic analytics with governance, evals, and an audit trail — in production, not a pilot. Open to AVP-level and senior platform leadership roles in financial services.

See the platform work →Résumé (PDF)
Williams — platform record, in productionJW-01Rev 2026-08Sheet 01
01Platform cost engineered out — Event Hubs → Kafka on Kubernetes$30K+/yr
02Quarterly reserve review — governed natural-language analytics via marketplace plugins3 days → 30 min
03Claude plugin marketplace — ~200 monthly users (40% of company), 2K+ queries/mo12 plugins
04Certified — AWS Solutions Architect + Cloud PractitionerAWS SAA

Numbers are measured, not aspirational — each one is expanded in a case study below.

02 · Selected work

Platforms in production

Written the way platform work actually happens: constraints first, numbers where they're real, diagrams instead of proprietary detail.

Flagship · AI platform

The Claude plugin marketplace

Spearheaded, built, and run the enterprise marketplace — 12 plugins and the authored domain skills behind them, spanning six business domains. Roughly 200 monthly active users — 40% of the company, five C-suite regulars — run 2K+ queries a month, at 30% lower token consumption than unassisted LLM use. I own the roadmap, the evaluations, the governance, and the support.

12 plugins · 6 domains

claudemcpai agentsllm evals
Case study →
AI & semantic analytics

Semantic analytics in production

A production Claude + Snowflake Cortex layer — Claims, Policy & Underwriting, Finance, and Legal ask governed questions in plain English — the quarterly reserve review went from 3 days to under 30 minutes; QBR prep from 4 days to under an hour.

QBR prep: 4 days → <1 hr

snowflake cortexdbtpython · sql
Case study →
Streaming platforms

The CDC platform rescue

Inherited a half-built replication platform, shipped it, then re-architected it — Azure Event Hubs → Apache Kafka on Kubernetes, with Debezium CDC connectors (SQL Server, PostgreSQL, Oracle LogMiner), custom Java connectors where none existed, and type-2 period-fact modeling for historical questions.

$30K+/yr saved

kafkadebeziumkubernetes
Case study →
Observability

Observability as code

Datadog and Prometheus dashboards shipped exclusively through GitHub Actions CI/CD, with forecast monitors that predict OOM and strain — the platform scales on signal instead of overprovisioning.

scale on signal

datadogprometheusgitops
Case study →
Open source — the work is public

Waybill — bring receipts

A Claude Code plugin for token accounting on AI-assisted work: deterministic attribution of agent token spend to shipped work, evidence tiers, conservation checks, and verification packs a recipient can re-run offline. The same discipline as the day job — every number traceable to a receipt — with the code in the open.

claude code pluginagent meteringMIT
github.com/Jakeintech/waybill ↗
03 · How I run it

Platform leadership, in practice

Own the number

Governance is the product. I run the BI tenant, so "governed" means the number is the same in the dashboard, the regulatory filing, and the chatbot — backed by SOC 2 SOPs and audit support.

Raise the level

Code reviews, mentoring, and office hours leading the org's adoption of AI-assisted development. The platform gets better when the people around it do.

Translate both ways

I serve every level of the business, analyst to executive — SME on feasibility, cost, and risk — including a key component of the enterprise pricing tool — a TypeScript microservice on Azure (GraphQL, MCP, PostgreSQL). Adoption decisions run on usage analytics, and model choices run on evaluations — evidence, not opinion, in both directions.

04 · Second act

Apps that hold up under load, too

Fitness & wellness for iOS — Swift, SwiftUI, HealthKit, on-device AI. Built by a runner and yoga practitioner who wants tools that work on-device and offline.

QuotedAI

Quotes that know your moment — an on-device engine reads time of day, energy, and activity via HealthKit to surface the right words. No ads, no trackers. 2K+ downloads.

SwiftUIHealthKiton-device AI
App Store ↗

App 02 — in flight

In development. It ships when it's confirmed on the App Store — not before.

in progress

App 03 — in flight

In development. The slot fills as the app clears review.

in progress
05 · Writing

From pipelines to agents

From Pipelines to Agents: What Governed AI Analytics Actually Takesoutline · in progress
The $30K Kafka Migration: Cost Engineering a Streaming Platformoutline · in progress
MCP in the Enterprise: Lessons from a Production Claude Pluginoutline · in progress
06 · About

Two tracks, one discipline

By day I build and run enterprise data platforms in specialty insurance — streaming, lakehouse, and the semantic layer that makes AI answers trustworthy. Nights and weekends I ship iOS apps. The through-line is the same everywhere: build things that hold up under load.

More about me →
07 · Direct answers

Who is Jake Williams?

A data & AI platform engineer at Vantage, a specialty insurer, where he owns the enterprise CDC/streaming platform (Apache Kafka on Kubernetes), leads engineering for a production Claude + Snowflake Cortex semantic-analytics system, and spearheaded the Claude plugin marketplace used across six business domains, analyst to C-suite. AWS certified; previously State Farm; author of the open-source Waybill plugin; indie iOS developer (QuotedAI).

What is the Claude plugin marketplace?

An enterprise marketplace of 12 Claude plugins and authored domain skills spanning six business domains. Business leaders at every level — including the C-suite — use it to build QBRs, reserve reviews, financial reports, and canonical dashboards from governed data. Jake created it and runs its roadmap, governance, and support.

What is semantic analytics?

Natural-language analytics built on an authored semantic layer — explicit definitions of metrics, domains, and joins — so an AI system answers from governed meaning rather than guessing at raw tables. Claude converses, Snowflake Cortex executes, and authored domain skills supply the semantics, keeping answers consistent with regulatory reporting.

Why does governed AI matter in insurance?

Because an ungoverned answer is a liability with good grammar. Insurance runs on regulated definitions and access boundaries; AI analytics only works there if it inherits them — consistent metrics, respected permissions, an audit trail. That governance-first pattern is what separates production GenAI from stalled pilots.

What does a data platform engineer do?

Builds and operates the infrastructure other data work stands on: ingestion and streaming (CDC, Kafka), the governed lakehouse (Snowflake, dbt, Dagster), orchestration, observability, and increasingly the semantic and AI layers. The product is the platform; its qualities are reliability, cost, governance, and trust.

Is Jake available for new roles?

Yes — open to AVP-level and senior data-platform leadership conversations in financial services. The fastest route is email (info@jakeawilliams.com) or LinkedIn (@jakeintech).

08 · The direct line

Talk data platforms.

Open to AVP-level and senior platform conversations in financial services — or to compare notes on governed AI analytics, Kafka cost engineering, and MCP in production. No forms, no calendar wall: a direct email gets a direct reply.

LinkedIn ↗GitHub ↗All contact options →