Led a team of engineers in building TKYO Drift, an open-source AI drift detection framework. I helped drive architecture decisions, worked hands-on with the backend drift analysis pipeline, and collaborated closely with the team across Python and JavaScript to ship a scalable, production-ready NPM package for AI observability — including support for embedding large datasets into vector space.

Selected and integrated a conceptual sentence-transformer model after evaluating multiple alternatives, balancing embedding dimensionality and inference latency to optimize vector similarity scoring for real-time drift detection and clustering workflows across millions of dynamic inputs.

Applied KMeans clustering on millions of high-dimensional embeddings to reduce vector storage by 60% and create semantically meaningful clusters, enabling faster similarity comparisons and more interpretable drift segmentation across large-scale inference workloads.

Integrated HNSW (Hierarchical Navigable Small World) indexing for approximate nearest neighbor searches on embedding vectors, reducing lookup times by over 90% to support scalable real-time drift detection and responsive semantic search across production inference pipelines.

Leveraged Python’s numerical computing stack (NumPy, scikit-learn, HNSWlib) to execute large-scale embedding generation, clustering, and approximate nearest neighbor search, achieving up to 50x faster processing compared to equivalent JavaScript implementations.

Architected a cross-language pipeline by spawning Python child processes from Node.js to handle embedding and similarity computations, enabling seamless integration between frontend services and high-performance backend models without blocking JavaScript’s event loop.

Refactored the AI drift detection pipeline into modular, class-based architecture to improve code organization, reusability, and traceability across complex vector workflows—reducing duplication, simplifying embedding logic.

Implemented client-side routing with React Router to enable seamless navigation across 10 distinct views, improving UX, reducing latency, and maintaining state without full page reloads.

Utilized MongoDB to store poll metadata and user credentials, leveraging its flexible document schema to support dynamic poll topics, rapid iteration, and scalable user management without rigid relational constraints.

Developed a Node.js and Express backend with modular routing and Mongoose models, enabling authenticated user registration, poll creation, and topic-based vote tracking through persistent, code-based room IDs.

Developed an interactive React interface enabling real-time comparisons between REST and GraphQL by allowing users to dynamically select return fields, visually highlighting GraphQL’s efficiency in minimizing overfetching and reducing payload sizes.

Designed a full-stack Node.js backend with Express and MongoDB, implementing modular routes, centralized error logging, and dual API support for REST and GraphQL; integrated real-time Pokémon API lookups and user CRUD operations within a unified data layer.

Implemented a flexible GraphQL schema using express-graphql to support declarative data access patterns, enabling nested user queries and field-specific Pokémon lookups; included resolver-level validations and schema-driven mutation logic for persistent data manipulation.

Operated advanced equipment including a 5' x 10' Armour CNC, 4' x 8' Saber CNC, Boss Laser 2K Fiber Laser Cutter, 5-axis CNC Foam Hot Wire Cutter, and 3D Scanner.

Interpreted designs and converted them into detailed assembly and fabrication drawings for CNC machining and 3D printing.

Led full-cycle project builds, from initial concept to final production, collaborating with both internal teams and clients.

Produced shop drawings for client approval, including projects requiring replication of existing designs.

Designed complex 3D parts using various software tools, preparing files for 3D printing or CNC milling.

Partnered with factory teams to fabricate intricate plaster designs.

Worked closely with clients and general contractors (GCs) to design practical, job-specific solutions.

Conducted site visits to assess field conditions and recommend fabrication approaches.