Hero Image

Ziggy Baker

Forward-Deployed Engineer

Applied AI & Product Engineering

I'm a forward-deployed engineer working where applied-AI research meets shipped product. I embed with teams to turn ambiguous, real-world problems into production systems — agentic workflows, multimodal generation pipelines, 3D and WebGL tooling, and the evaluation harnesses that keep them reliable.

Today I lead Applied AI & Product Engineering at Orbital Vision. Before that I founded and ran Petal, a florist-native, multi-tenant alternative to Shopify.

claude / anthropic api · python · typescript · next.js · react three fiber · webgl2 · comfyui · stable diffusion 3.5 · controlnet · aws (sqs · dynamodb) · firebase · postgres · clickhouse · stripe · embeddings · docker · git

Contact Me

Orbital Vision

Nov 2024 – Present

Technical Director — Applied AI & Product Engineering

I joined Orbital Vision as an Applied AI engineer training image-generation models, then moved into technical leadership of OV25 — Orbital Vision's AI, 3D configurator, commerce and developer-platform product. The throughline of my work there is taking historically manual, expert-only processes and turning them into safe, agentic, AI-assisted systems. A breakdown of everything I've built and led:

Applied AI & generative imaging

  • Post-trained an internal asset-generation model on Stable Diffusion 3.5 — depth/Canny ControlNets, prompt conditioning, reference-image control, denoise tuning, and guardrailed review loops for product-image quality.
  • Built AI material and texture generation trained on Orbital Vision's 10-year V-Ray archive, translating high-fidelity offline-render material knowledge into WebGL-compatible PBR textures and real-time browser configurator materials.

Evaluation & generative quality

  • Built automated evaluation frameworks running thousands of generated variants across ControlNet strength, denoise steps, sampling algorithms, prompt structures, and reference-image strategies — with model-based scoring of visual fidelity, product accuracy, brand alignment, artifacts, and commercial usability.

Agentic 3D-configurator generation

  • Led AI-assisted creation of web-based 3D configurators — automating a historically manual process complicated by inconsistent model structure, mesh naming, material setup, pricing rules, and customer-specific logic.
  • Designed agentic setup patterns where users reference @materials, @products, @pricing and @configurator in natural language to generate material mappings, visibility logic, option dependencies and pricing conditions — all with review-before-apply diffs.
  • Replaced complex UI that required heavy staff training with plain-language commands backed by sandboxed, safe agentic skills and tools.

Multimodal 3D understanding

  • Built a multimodal 3D asset-understanding pipeline that renders geometry, vertex data, and mesh-level components from multiple viewpoints, using Meta vision models and Claude Opus to infer component purpose, material groupings, and product semantics for AI-assisted configurator generation.

High-resolution WebGL capture & automated QA

  • Built a custom WebGL/MSAA capture pipeline rendering live React Three Fiber scenes into offscreen WebGL2 render targets at 4K/8K with GPU-aware multisampling.
  • Used those render pipelines to automate 3D QA — detecting incorrect UV mapping, floating, colliding or misaligned meshes, and other visual imperfections — saving hundreds of hours.
  • Hardened the capture system for operational safety under heavy workloads: persistent browser pools, CDP-driven memory cleanup, worker queues, retry budgets, screenshot fallbacks, graceful shutdown, SQS visibility extension, and recoverable task state in DynamoDB.

The company “brain”

  • Centralised a decade of company knowledge — scattered across NAS drives, cloud storage, email accounts and internal notes — into a single indexed, embedded, searchable library, reusing the render pipeline to convert 3D files into rendered images, embed them, and make roomsets, models, and assets discoverable.
  • Designed an agentic brief → retrieval → generation workflow: take a client brief, find a close-match roomset or model in the company brain, generate any missing reference accessories on the fly (image → multi-view synthesis → LRM → surface extraction → mesh/texturing, inspired by NVIDIA research), and hand the close match plus prebuilt props to 3D artists — halving their work.

Production AI platform

  • Designed and shipped production multimodal AI workflows across Claude/Anthropic, Gemini, Veo, OpenAI, Meta vision models, Flux and ComfyUI — async job orchestration, prompt expansion, retry/failover handling, token ledgers, refunds, safety checks, and usage analytics.

Technical leadership

  • Led a team of engineers across OV25 — architecture direction, pair programming, implementation support, and technical review — turning repeated implementation problems into stronger, reusable platform abstractions.
Read the full Orbital Vision breakdown →

Petal

2021 – 2026

Founder / CTO

I founded and led Petal, a florist-native, multi-tenant alternative to Shopify — storefronts, CMS, analytics, inventory, promotions, regions, localisation/i18n, delivery rules, subscriptions, ticketed workshops, media, domains, tenant Stripe checkout, onboarding, and operational dashboards. Alongside the commerce platform I prototyped agentic marketing systems that model each florist's catalogue, delivery zones, seasonality, and brand voice as connected context for AI-assisted sales and SEO.

Earlier projects · 2023

© 2026 Ziggy Baker