Built the Marsha AI Programming Language* Simple, markdown-based syntax to describe what a function should do with a few examples of usage to generate working Python code.* Generates code and a test suite for said code, and iterates with the test suite and an LLM until the tests passes.

Built IaSQL - Managing your Cloud Infrastructure via SQL* Infrastructure as Data - AWS infrastructure represented as PostgreSQL tables and relations/dependencies as simple foreign key joins* Two-way Source of Truth - By creating an idempotent representation of the cloud and database information, coupled with an audit log of database changes, mutations of cloud infrastructure outside of IaSQL are properly imported into the Postgres database while changes by users in the Postgres database can be simultaneously executed back into the cloud, providing a resiliency beyond Terraform.* Easy import - due to the above feature, you can connect to any existing AWS account and immediately begin managing it, no need to write any source code first.

Built a beta form of the Alan Programming Language - an implicitly parallel, safe programming language* Syntax inspired by Typescript, type system inspired by Rust, exploring the limits of a barely-Turing-incomplete language with the goals of automatically inferring all types at compile time (almost completed) and preventing many runtime errors like infinite loops (completed), integer overflows (completed), buffer overflows (completed), and parallelization/clustering decision making (many advances listed below).* Static analysis to determine an exact dependency graph of source code, encoded in custom bytecode format, to give runtime the ability to automatically parallelize execution where it makes sense. Primarily used to segment IO-bound opcodes from CPU-bound, parallelizing the IO-bound operations.* Built-in clustering for transparent IPC between nodes, autoscaling by load across multiple nodes in one or more region across AWS/GCP/Azure (all of them simultaneously if desired), and a built-in rendezvoushash-based distributed hash table "datastore" (with region-aware replication and routing) that could be used similarly to memcache or (because we control the language and runtime) a distributed compute mode that would determine if it is worth it to transport the thread of execution (and its closure scope) to the node with the data or pull the data to the current node to perform the execution to automatically minimize latency.While we demonstrated that you could have an ergonomic language that automatically parallelizes and distributes your code, confirms type safety without (most) annotations, serialization into JSON without Turing completeness, and got our beta runtime faster than CPython in testing, we proved (to ourselves, at least) that you cannot parse arbitrary data without Turing completeness, so we have shelved the language for the time being until/unless we can add limited Turing completeness without compromising automatic parallelization.

Worked primarily on backend and devops projects at CarDash.* Implemented many backend (and a couple of web frontend) services in support of exploratory efforts at CarDash.* Implemented automated pricing algorithms for car repair procedures based on historical data and real-time parts pricing.* Implemented scheduled maintenance procedure lookups to accelerate and automate estimates for customers.* Improved developer speed and eliminated some architectural antipatterns by switching to microservices (the trade-off was net positive, that is).* Reduced datacenter spend by replacing Elastic Beanstalk managed backends with a classic Bastion + Server Pool deployment system.* Introduced an RFC procedure on the team for large or contentious changes, along with a template to make sure all major technical and business concerns are addressed.* Improved monitoring and on-call to catch outages more quickly. Sub-1-minute response time and sub-5-minute recovery time the target (and sometimes reality).

Worked on many parts of the core of Uber:The core dispatching system for over a year primarily on scaling and integration of new features between app and backend services.The "Share My ETA" feature for a few months primarily on its data ingest architecture (ported to Kafka from Redis Pubsub).Part of the original UberPool team for 6 months refactoring the dispatching system to handle multiple clients on the same trip and integrating the matching algorithms.Supply forecasting and positioning for about 4 months working primarily on data processing (kafka and HTTP sources).The surge pricing team for a year and a half architecting the new surge pipeline, helping develop 2 new surge algorithms, including the new UberEats surge, and helping write a new spatial indexing algorithm in C (H3: https://github.com/uber/h3).The UberEATS pricing team for 9 months rearchitecting the various pricing systems into a unified, horizontally-scalable, 4-layer stack to improve reliability, monitoring, and experimentation. This allows the team to accelerate engineering and data science efforts with confidence that their efforts won't break other parallel efforts with respect to price within the EATS org.

Research work on Gate Oxide Breakdown and Electrostatic Discharge

Developed internal applications using MS SQL Server, ColdFusion, and Javascript.

First developer and chief architect of technology stack for a friend's stealth startup. Consists of a collection of services accessed via JSON-RPC. These services are Node.js front-ends to MongoDB databases, Solr search engines, ImageMagick, and various Amazon AWS services (EC2, S3, SES).This is coalesced with an Express-based Node.js service that runs the actual website, which is designed to be stateless (state pushed into session store) so multiple can be spun up and reversed proxied (NginX or Lighttpd, currently Lighttpd but looking at transition). The Express-based service uses jQuery extensively for interactive elements as well as the views with jQuery Templates.I wrote several supporting libraries (most open-source and visible below, some not) in support of this, but have only been able to work in spare time because it's an unfunded startup.

ESD Test Chip Layout Automation