* Architected and developed "Dollhouse" using Three.js, projecting 2D floorplans into 3D objects

* Improved AskDigs' RAG pipeline, decreasing average response latency from 10s to 2s

* Added new prediction capabilities (e.g. orientation) to open-source models, unlocking new use-cases with minimal investment

* Integrated AI predictions into takeoffs, allowing homebuilders to quickly estimate flooring quantities

* Used AWS & BentoML to deploy multiple computer vision models powering realtime and batched inferences

* Used Azure ML Studio & HuggingFace to train and fine-tune computer vision models

* Merged fixes for several upstream open-source libraries, including HuggingFace Transformers, Drei & Hasura

* Developed and trained custom PyTorch image models to parse out key information in floor plans (rooms, structures, etc.)

* Designed Terraform configurations through CDKTF to spin up and manage cloud infrastructure with multiple providers

* Launched an internal analytics program with Metabase that non-engineers used to answer data questions

* Created patented algorithms around easing WebAR user interactions

* Translated company metrics into consumable data definitions in Looker

* Experimented with alpha WebAR technology for new company use cases

* Worked with cross-device mesh data as both a visualization and user-interaction tool

* Built out custom analytic databases for key customers

* Created new use cases for the core web experience

* Overhauled analytics infrastructure and pipeline from generation to analysis

* Generated personalized usage reports for high-value customers

* Redesigned cross-platform augmented reality systems to sync 3D data (pose, world data, elements, etc.) across clients

* Developed Kotlin server endpoints to perform operations on a PostgreSQL database