i’m gabe, a full-stack engineer and product builder with experience across low-level security research, customer-facing architecture, independent products, and applied ai systems.
across those environments, a recurring pattern has been taking ambiguous problems from early exploration to working software: learning unfamiliar systems, combining existing tools and technical capabilities, finding useful abstractions, and adapting the implementation as the real constraints become clearer.
since january 2026, i’ve been a senior full-stack engineer at Meshy AI. my work has included ai application generation, coding-agent infrastructure, content moderation, and agentic workflows for internal teams.
outside of building and working, i’m likely hanging with my wife either eating sushi, ballroom dancing, or tiring out our mini aussie.
my strongest pattern is taking messy inputs and finding the abstraction that makes the problem easier to reason about.
user feedback, domain constraints, adjacent tools, technical limitations, and random conversations all tend to mix together until a simpler representation of the problem starts to emerge. once that representation becomes clear, the work has often moved through a fast initial build, testing against users or system behavior, and deeper redesign as new constraints appear.
blocksmith is the clearest example. it started as an image-to-voxel prototype because that seemed like the obvious path. feedback from hytopia developers made it clear that voxel models were not actually the right representation for minecraft/hytopia-style assets. that pushed me to rethink the problem around llm-generated semantic block entities instead. later, when raw json generation became the bottleneck for complex models, i replaced it with a pythonic dsl that could compile into the same model formats while letting the ai use loops and structure.
that pattern has repeated throughout my career: security research at raytheon, marketplace/product architecture at tabu art, customer workflow mapping at stripe, ai product building with blocksmith, and now product systems work at meshy.
this does not mean every problem needs a new abstraction or a greenfield build. often the right answer is understanding what already exists, reusing it intelligently, and adding only the missing layer.
blocksmith began after i encountered an asset-creation bottleneck while building games for hytopia. the first prototype produced voxelized models, but developer feedback showed that the platform needed editable, low-complexity models made from semantically meaningful cuboid parts.
that led to an llm-to-dsl generation system that compiled models into glb and blockbench-compatible formats, along with a separate multimodal texturing pipeline.
the product gained free and paid users, including teams that created hundreds of game assets. i later sunset the closed saas after the market, pricing, training-data constraints, and investment required for the next quality leap no longer justified continuing it as a company. the core generation engine is now open source.
model-generation case study · open-source core · texturing writeup
at meshy, my work has included: