Kylan Thomson AI Engineer & Google Cloud Solutions Architect

Where I've worked

Professional Experience

1. Google Cloud Solutions Engineer · GenAI · Promevo

   Sep 2025 – Present

   Design and deliver enterprise GenAI and agentic solutions on Google Cloud for Fortune 500 clients, from AI strategy and value discovery through architecture and production deployment.

   • Built Promevo's No-Code Agent Library, an AI value-discovery platform generating 100+ role-specific agent use cases across 10 business functions with ROI modeling, surfacing an average of $3M in agentic opportunities per engagement.
   • Helped position Promevo as a strategic Google Cloud GenAI partner via the Agent Library and a repeatable Gemini Enterprise methodology, contributing to Google entrusting Promevo with $1M–$2M enterprise opportunities.
   • Architected a reusable suite of "last-mile" agents (Python, Google ADK, Agent2Agent) that autonomously produce presentations, images, video, and documents grounded in centrally managed brand guidelines and templates.
   • Exposed those agents over A2A as composable enterprise infrastructure: customer agents delegate artifact creation while a central control plane governs brand standards and approved assets.
   • Built a multi-MCP marketing and commerce agent for a global digital experience platform that unites audience intelligence, product insights, generative video, Meta ads, and Shopify campaigns in one workflow, winning C-suite sponsorship for production and marketplace distribution.
   • Led an agentic AI hackathon for a major U.S. manufacturer, turning supply-chain challenges across ~12 teams into an orchestrated agent that surfaces forecasting insights conversationally and enables earlier intervention against disruptions, multimillion-dollar expedited-shipping exposure, and plant-shutdown risk.
   • Productized Gemini Enterprise delivery (workshops, use-case discovery, ROI analysis, change management) so engagements repeat across Promevo's Professional Services organization.

   • Vertex AI
   • Gemini Enterprise
   • Agent Development Kit (ADK)
   • Agent2Agent (A2A)
   • MCP
   • Python
   • Google Cloud

2. Experimentation Specialist · Alaska Airlines

   Oct 2023 – Sep 2024

   Brought LLM automation and large-scale data engineering to the experimentation program of a major U.S. airline.

   • Led development of LLM-powered agentic workflows that automated experiment analysis end to end, with human-in-the-loop validation so analysts reviewed conclusions instead of rebuilding them, an early production application of GenAI inside the experimentation program.
   • Engineered Python/PySpark pipelines on Databricks that evaluated experimentation techniques at scale, giving product teams a defensible, data-driven basis for ship-or-hold decisions.
   • Deployed a Spark-based QA pipeline that inspects 100,000+ transactions a day for anomalies and system-health signals, giving the program a continuous check on the integrity of the data behind every readout.
   • Built experiment tagging and tracking systems that turned a growing portfolio of tests into actionable insight on customer behavior and revenue impact.
   • Designed and ran the experiments behind the airline's eco-friendly initiatives, lifting user engagement and conversion on those flows.

   • Python
   • PySpark
   • Databricks
   • LLM Workflows
   • Spark

3. Software Engineer · Credera

   Nov 2021 – May 2023

   Consulting engineer delivering cloud data platforms and full-stack products for enterprise clients.

   • Built real-time customer data streams on AWS MSK (Kafka) feeding Braze, enabling personalized, event-driven customer engagement for an enterprise client.
   • Implemented infrastructure as code with Terraform and GitHub Actions CI/CD pipelines, making environments reproducible and every deployment an automated, reviewable change.
   • Wrote Java integration test suites that guarded pipeline integrity as the data platform scaled.
   • Led website localization with React and Node.js, increasing global user engagement across the client's international audience.

   • AWS MSK / Kafka
   • Terraform
   • Java
   • GitHub Actions
   • React
   • Node.js

4. Software Engineer · UNO

   Oct 2020 – Nov 2021

   Full-stack engineer at an early-stage startup, shipping across web, mobile, and cloud.

   • Delivered scalable product features across the full stack (React and Node.js on AWS, Python services, and Flutter mobile) with ownership from design through deployment.
   • Turned business requirements into technical specifications that kept a fast-moving roadmap aligned with stakeholder goals.
   • Profiled and optimized application performance and AWS backend operations, improving responsiveness and reliability.
   • Developed cross-platform iOS and Android apps with Flutter from a single codebase.

   • AWS
   • React
   • Node.js
   • Python
   • Flutter

What I've built

AI & Cloud Projects

SporeCast

Geospatial machine learning that predicts where edible mushrooms are fruiting across Washington State — species distribution modeling over satellite embeddings, terrain, and weather-aware phenology.

20 species modeled from 13 public data sources, served as an interactive forecast map.

• Built a per-species ensemble distribution model under spatially blocked cross-validation, calibrated so scores stay comparable across species.
• Engineered 40+ covariates spanning 64-dimensional satellite embeddings, terrain, hydrology, and lapse-rate-downscaled climate normals.
• Corrected presence-only observer bias so the model learns habitat quality rather than where people happen to hike.
• Shipped the whole pipeline on keyless public APIs and server-side reductions — no raster downloads, fully reproducible.

• Python
• scikit-learn
• Google Earth Engine
• LightGBM
• Leaflet
• Cloud Run

Read the technical case study→

Pokémon TCG Live Attacks

A playable digital Pokémon Trading Card Game where a deterministic rules engine decides every attack first and a generative video pipeline animates exactly that outcome — 20,444 cards, online play, and a Gemini director, all on Google Cloud.

20,444 cards across 174 sets playable, 67% of 27,927 printed attacks resolved exactly by the engine, and every attack animated from its decided outcome.

• Built a pure, seed-driven rules engine — evolution, special conditions, multi-prize knockouts, modern and classic rule sets — where printed card text is parsed into structure and a matcher may only claim a sentence it has read in full.
• Designed a generation pipeline that resolves the attack first, then composes keyframes and a prompt from the decided outcome, so the video describes what the board will do rather than the other way round.
• Cut generation latency with cache warming: scene portraits, a Gemini-written direction, and the likely opening attack are all generated the moment a card reaches a hand, turns before they are needed.
• Shipped server-authoritative online play over SSE with per-seat redaction, a Firestore match document any Cloud Run instance can serve, and secrets mounted from Secret Manager with the director authenticating as the service account.

• TypeScript
• Next.js
• Cloud Run
• Firestore
• Gemini
• fal.ai

Read the technical case study→

Resume Revival

AI-powered web application that optimizes resumes and cover letters for ATS systems, with multi-job tailoring and keyword optimization.

6,000+ job seekers served, improving interview chances through keyword-optimized resumes.

• Architected and shipped a production GenAI application on Google Cloud Functions with Gemini integration.
• Engineered an AI pipeline that generates keyword-rich, ATS-friendly resumes and cover letters tailored to specific job descriptions.
• Built multi-job tailoring and skill-gap analysis features that show candidates exactly what to develop for their target role.

• Google Cloud Functions
• Gemini
• React
• Python
• API Development

Wildfire Sentinel

A live-data wildfire simulator and response platform: pick any place on Earth and any historical date range, watch a physically-grounded fire fight real terrain, buildings, and weather — then race autonomous drone doctrines to find what would have saved the most.

Reproduces the 2025 Palisades fire to within order-of-magnitude on structures lost with no per-fire tuning — from 4 live public data sources and zero API keys.

• Rebuilt a hackathon prototype into a self-contained simulator that assembles any scenario on demand from four key-less public APIs — 30 m elevation, 45,930 OpenStreetMap structures, water polygons, and hourly ERA5 weather — with no backend and a fallback for every source.
• Replaced probabilistic cell flips with a physical spread engine: rate of spread in metres per hour shaped by an elliptical wind field, lagged fuel moisture, slope, and ember spotting, which is what lets the model be validated against a real fire rather than tuned to one.
• Built a Strategy Lab that clones a paused incident and races response doctrines against seeded-identical fires, so outcome differences are attributable to strategy rather than luck, and the winning algorithm can take over the live simulation in one click.
• Added a Monte Carlo damage forecaster projecting burn probability and value-weighted expected loss 24–72 hours ahead, chunked through the event loop to keep the interface responsive.

• React
• JavaScript
• Canvas
• Three.js
• Monte Carlo
• Geospatial APIs

Read the technical case study→

AI Stock Image Factory

A demand-driven generative pipeline that reads search-volume data, generates stock imagery aimed at what buyers actually search for, and engineers the metadata that makes it findable.

5,110 downloads and $3,840 in royalties earned on Adobe Stock, from a fully automated generation pipeline.

• Built an end-to-end pipeline that samples target keywords weighted by monthly search volume, so production tracks real buyer demand rather than working a list top-down.
• Engineered the metadata path that actually drives sales: model-generated keywords are deduplicated, re-ranked by embedding similarity to the target term, and filtered, so the highest-intent terms land in the slots the marketplace weights most.
• Layered a library of 72 named house styles over fixed core instructions, keeping output visually varied but consistent in quality — and model-agnostic enough to swap image models as a config change.
• Made the model stack fully config-driven with per-stage tiering, holding cost to a few cents per asset while reserving spend for image generation.

• Python
• Generative AI
• Prompt Engineering
• Embeddings
• Content Ops
• Cloud Run

Read the technical case study→

Notes from the work

Writing

Technical pieces that extend the case studies: how a rules engine keeps a video model honest, why random cross-validation flatters a spatial model, what paired random seeds do for a simulation. Each one is grounded in a system I built and ran.

• Geospatial ML · Forecasting

  Machine Learning Mushroom Forecasting for Washington State

  September 24, 2026 · 9 min read

• Geospatial ML · Data Quality

  Correcting Observer Bias in Citizen-Science ML Datasets

  September 21, 2026 · 9 min read

• Generative AI · Content Operations

  Building AI Image Pipelines Around Search Demand Instead of Prompt Quality

  September 21, 2026 · 9 min read

What I work with

Skills & Technologies

Credentials

Certifications & Education

Google Cloud Certifications

Google Cloud Generative AI Leader

Issued Oct 2025

Validates expertise in designing and deploying cutting-edge generative AI solutions on Google Cloud.

Google Cloud Professional Data Engineer

Issued Apr 2022

Demonstrates expert ability to design, build, and manage data processing systems on Google Cloud for reliability and scale.

Google Cloud Associate Cloud Engineer

Issued Apr 2022

Confirms foundational knowledge of deploying applications, monitoring operations, and maintaining projects on Google Cloud.

Education

Virginia Commonwealth University

Graduated Mar 2021

B.S. Computer Science, Minor in Statistics

Dean's List. Founded CS Certs, a student organization funding certification study materials and exams for web development and computer science.

Get in touch

Let's Build Something on Google Cloud