Collect and Generate Data
Before anything can be predicted or visualized, it has to exist and be trustworthy. This is the acquisition layer: instrumenting sources, capturing telemetry and video, generating structured data where there was none, and landing all of it on a clean, production-ready foundation.
Data Engineering & Integration
Designing high-throughput ETL/ELT pipelines, structured warehousing layers, and database migrations that establish a clean, production-ready data foundation.
This is direct, not theoretical: it's the same discipline that migrated Rakuten's telemetry from 40M+ monthly active users to BigQuery, moved Nippon TV's Factly platform off Treasure Data, and structured nationwide mobile network telemetry at NTT Docomo.
Cloud-Native Architecture
High-integrity serverless and containerized environments built on GCP and AWS. Engineered to manage ingestion pipelines, real-time telemetry, and microservices.
Proven at Rakuten's full GCP migration and at P&I Information Engineering, where a legacy Java/C estate was rebuilt on modern PHP and Terraform-managed AWS infrastructure.
Data Security & Protection
With over 15 years defending enterprise data in production, security isn't bolted on after the fact — it's engineered in from day one. From encrypted pipelines processing transaction telemetry from 40M+ monthly active users at Rakuten to the IAM-hardened auth gateway built for GalaSpo's own streaming platform, every dataset we touch is treated as a live threat surface, not just plumbing.
Manage, Predict and Visualize Data
Once the foundation is solid, this is where the data earns its keep: governed and warehoused, modelled for prediction, driven by autonomous agents, and turned into interfaces people actually reach for.
Data Science & Analytics
Insights and statistics can now be visualized in ways that were simply unimaginable a few years ago, powered by AI systems running on LLMs. Outputs that used to need a dedicated engineering team and months of work can now be built as a tailored, purpose-built application in under an hour.
We treat apps the way you'd treat a disposable ziplock bag from here on: built for one use, then thrown away, not maintained forever as a permanent product. We can lead your team into that same workflow.
The same instinct behind shipping TelegraTalk and Simple Trump Card in days, not months, and the statistical modelling behind predictive systems at NTT Docomo, Sompo, and Rakuten's Marketing View Premium dashboard.
Agentic AI Integration
Orchestrating autonomous workflows and multi-agent loops that run asynchronously to handle enterprise project planning, code synthesis, and analytics.
Grounded in building Model Context Protocol client tools at Jitera and Rakuten, both bridging LLMs to real, governed enterprise data rather than a demo sandbox.
Leadership & Track Record
15+ years of software engineering, cloud architecture, and AI integration for global enterprises — from a first job in Richmond, BC to leading AI strategy across Tokyo's biggest tech companies.
William Joe Baldwin
I'm the principal architect behind Baldwin Data Works: over 15 years turning complex legacy workloads into cloud-native systems, turning tens of millions of consumers' transaction telemetry into analytics platforms for enterprise B2B retail clients, and now architecting agentic AI platforms. What I bring to a team isn't just technical range, it's the flexibility to start a large project completely from scratch and the discipline to run it the traditional, battle-tested way when that's what the job actually calls for.
Principal Product Owner & AI Technical Lead
Jitera | Remote (Vancouver, BC)
Directed architecture and delivery of enterprise apps using advanced RAG frameworks, and implemented Model Context Protocol (MCP) client tools bridging LLMs with structured enterprise data sources. Standardized task workflows across Agile GCP deployments, mandating that every ticket define the Why, What, and How before work began.
Lead Product Manager & Technical AI Lead
Rakuten | Tokyo, Japan
Led 19-member engineering teams combining transaction telemetry from 40M monthly active users across Rakuten's 6-million-client retail network. Pioneered the full migration to a GCP/BigQuery-native architecture, engineered Marketing View Premium (a B2B analytics dashboard for 200+ supermarket chains across Japan), and helped pioneer early MCP tooling as part of Rakuten's AI-nization movement.
Data Consultant & Senior Database Developer
SoftBank Group Corp | Tokyo, Japan (on-site, via Gruff Inc.)
Designed and optimized complex data architectures for massive-scale telecommunications workflows, and delivered clear roadmaps for migrating legacy on-premises infrastructure to modern cloud environments.
Project Leader & Machine Learning Engineer
Nippon TV | Tokyo, Japan (on-site, via Gruff Inc.)
Led a 9-person team migrating a massive data pipeline from Treasure Data to GCP BigQuery for the Factly platform, and integrated ML models including celebrity face recognition and video genre categorization for broadcast applications.
Lead Data Scientist & Machine Learning Engineer
NTT Docomo Inc | Tokyo, Japan (on-site, via Gruff Inc.)
Led a 12-member data analysis team extracting insights from nationwide mobile network telemetry, and deployed predictive systems including a loyal-customer-likelihood model and a subscriber-candidate predictor.
Project Leader & Data Scientist
Sompo Holdings | Tokyo, Japan (on-site, via Gruff Inc.)
Developed machine learning solutions for the insurance sector, including a predictive car-insurance cost model and a customer churn prediction model.
Project Development Manager & Lead Data Scientist
Gruff Inc | Tokyo, Japan
The consulting entity behind the SoftBank, Nippon TV, NTT Docomo, and Sompo engagements above, managing simultaneous delivery of cloud architectures, Data Management Platforms, and big data pipelines across AWS, GCP, and Azure for major telecom and broadcasting clients. Running four concurrent client engagements at once is what turned disciplined ticket and task management into the core of how I still run projects today.
Technical Lead & Software Engineer
P&I Information Engineering Co Ltd | Tokyo, Japan
Architected a diverse portfolio of enterprise systems, including a cloud-based analytics platform and a Pepper robot reception system with natural-language interfaces. Spearheaded the migration from legacy Java/C into modern PHP and cloud architecture on AWS/Terraform, and led development of a radioactive-soil traceability database adapted from the company's own automotive Bill-of-Materials logistics logic. This is where my approach to architecture was really forged, before I carried it into Gruff, Rakuten, and Jitera.
Technical Engineer
Azur Productions Inc | Tokyo, Japan
Engineered internal tools including a flex-time stamping system and a proprietary video file-sharing server to streamline post-production workflow efficiency, hands-on video tooling experience that later shaped GalaSpo's own streaming architecture.
Independent Web Developer
Self-Employed | Tokyo, Japan
Programmed and operated numerous web platforms using PHP, JavaScript, MySQL, and Google App Engine, including a marriage-agency API and a salon reservation system.
Web Designer
Canada @ Home Inc | Richmond, BC, Canada
Managed exporting Canadian products to the Japanese market, and designed web pages and internal tools with HTML, JavaScript, Excel, and Access. My first job, and where I first started applying database-driven systems thinking I'd already been doing informally since my teens.
Recognition
Solidarity Grand Prize
Rakuten — awarded for succeeding as a cross-divisional team.
Most Valuable Product Award
Rakuten Technology Division.
Conference Speaker
Itochu's internal corporate conference, Dallas, TX (Feb 2026) — “What is an AI Context Platform.”
Active R&D and Intellectual Property
Proprietary prototypes demonstrating our technical leadership in high-integrity software engineering — grouped by the two halves of the data lifecycle above.
01 Collect and Generate Data
GalaSpo
A highly modular, microservices-based monorepo platform designed for live-streaming high-fidelity sports events. Utilizes a security gateway proxy, real-time metadata coordination, and camera-first mobile RTMP integration. The video codec and bitrate-adaptive encoding pipeline draws directly on hands-on post-production tooling built at Azur Productions and broadcast-grade video ML work delivered on-site at Nippon TV.
- React Native
- Docker & Nginx
- Auth Gateway (IAM)
- RTMP Streaming
GalaSpo Stats
The generation half of the GalaSpo platform: an iOS app that turns a plain recorded match into structured data. It tracks every player, the ball, and the camera frame by frame — numbered head pins tinted to shirt colour, a constant-velocity predictor bridging occlusions, and the camera solved as a fixed touchline rig from the pitch lines — then writes the whole match to a per-session SQLite trajectory database in real pitch metres, plus a GeoJSON export that loads straight into kepler.gl or QGIS.
It reads the machine-readable camera-orientation strip that GalaSpo Football Streamer burns into its recordings, so the two apps compose: Stream collects the footage, Stats generates the dataset. The recorded frames, detections, and self-corrections are laid out as training data, with a CoreML seam that lets a stronger model drop in and the on-device heuristics stand down — the same broadcast-video ML lineage as the on-site work at Nippon TV.
- SwiftUI
- Apple Vision
- Core ML
- SQLite / GeoJSON
- Homography & Camera Pose
FMify
A continuous internet radio player for iOS with a live DJ narration layer — equal-power crossfading between two decks, on-device vocal/BPM detection, and a 100-track program builder that schedules a target vocal ratio, BPM cycling, and repeat/artist cooldowns before a single track plays.
The narration timing is where this one gets personal: it's built directly on my own 4.5 years hosting live FM radio in Nagoya, Japan — the same discipline of finishing a spoken intro in the exact instant the vocals begin, now running as a scheduling algorithm instead of a headphone cue.
- SwiftUI
- AVAudioEngine
- WeatherKit
- NewsAPI.org
terranian
Pick any point on Earth and it generates a theoretical 3D world for that location on the fly, live in the browser — real terrain elevation, extruded buildings, roads, forests, farmland, and water, all reconstructed from OpenStreetMap and public elevation data at generation time, no pre-baked assets.
A natural extension of the geo-analytics and BI visualization work in the BI Tool Suite, pushed into procedural 3D: coastal water, for instance, isn't tagged as a fillable area in OpenStreetMap at all — it's reconstructed from raw coastline boundary lines using a grid-sampled nearest-boundary classifier and marching squares, verified against real map tiles rather than eyeballed. The data layer is kept deliberately renderer-agnostic, so the same pipeline is designed to eventually drive an Unreal Engine target as well as the browser.
Two intended directions: a geographic analytics surface for exploring real spatial data in 3D instead of on a flat map, and a procedural field generator for video games, where a real location's layout seeds a playable level instead of a hand-built one. The visuals are intentionally minimal right now — this is the data pipeline proving itself out first — and will keep improving as the rendering layer matures.
- React Three Fiber
- TypeScript + Vite
- OpenStreetMap Overpass API
- Mapbox Terrain-RGB
02 Manage, Predict and Visualize Data
Gomaae (γγΎεγ)
A desktop project management platform structured around Japanese HOU REN SOU (reporting, communication, consultation) principles. Leverages custom Agentic AI workflows to streamline the SDLC and bridge teams with structured data layers.
Still R&D, but the idea is straightforward: manage a project the way you would with a real team, except your teammates can be AI companions too. HOU REN SOU is baked in for a reason — after 20 years navigating exactly that reporting culture at Rakuten, NTT Docomo, and SoftBank, I've seen firsthand where communication actually breaks down on a distributed team.
- Tauri v2 (Rust)
- Next.js Standalone
- SQLite
- Anthropic Claude API
TelegraTalk
A hands-free, voice-only Telegram client. Incoming messages are read aloud automatically and replies are captured by speech — the entire conversation controlled by a single click of a headset button.
No specific past role ties to this one — I built it purely out of wanting to talk to AI agents connected to Telegram while I'm on the go: walking, jogging, driving. It earns its place here as an interface experiment: structured message data delivered entirely by voice.
- SwiftUI
- TDLibKit
- Speech Framework
- Apple Intelligence
Simple Trump Card
A physical deck of cards, digitized — draw, flip, pass, and pile a shuffled 54-card deck exactly like you would at a real table, with no rules or scoring baked in. Supports both local pass-and-play and peer-to-peer multiplayer directly between nearby iPhones over Bluetooth/Wi-Fi, with no server or accounts involved.
No particular past experience behind this one either — it's a demonstration of shipping a genuinely unique interface in just a few days with LLM-assisted development.
- SwiftUI
- MultipeerConnectivity
- XcodeGen
My Philosophy
Three principles, each one earned on the job, not written in a boardroom.
Extreme Systemization
I've been building database-managed systems since I was a teenager, and I put that instinct to work in my very first job at Canada @ Home. It was my years at P&I Information Engineering, though, that really polished it into an architectural discipline, one I carried straight into Gruff, Rakuten, and Jitera: automate the routine, systematize the complex, and push operational efficiency to its limit.
Intentional Engineering
At Gruff, I was leading four different projects across four different client organizations at once. Systemization was already my default way of solving problems, and running that many engagements in parallel is what turned disciplined ticket and task management into the system that actually keeps it all in harmony. Every line of code, and every ticket, still has to answer the Why, What, and How before I touch it.
A Culture-First Future
Long before ChatGPT's breakthrough, a colleague and I used to spend our downtime debating the 2045 singularity. I still believe AI will take over most of our tasks, but that doesn't mean people are left with nothing to do. What's left narrows down to culture, and I think that's the prime motivator in life, right behind money.
Corporate Business Card
Interact with our executive business card. Hover to tilt in 3D, click to flip, or download a printable layout.
William Joe Baldwin
Chief Executive Officer
BALDWIN DATA WORKS INC.
Incorporation Number: BC1552899
OPEN TO OPPORTUNITIES
Build with Us
Discuss C2C partnerships, scalable cloud-native architectures, or high-performance Agentic systems.
Currently Open to Opportunities
Actively seeking new partnerships, contract roles, and impactful projects.