I'm an avid pythonista who sometimes finds himself writing Rust, Typescript or Go. I don't work in tech and don't develop professionally, but I aim to write professional code. If my code doesn't meet profesional standards, please tell me; Since I don't work in tech, I can sometimes overlook the obvious approach. On the plus side, I occasionally stumble into novel approaches too.

My work here is just an extension of my obsessive need to solve interesting problems and in the process, go way (way) too deep. The problem solving process is what I enjoy the most, at least in my personal life.

My interests are diverse and are always changing. They include reinvisioning ideas from an AI/ML-first perspective, data science, cryptography, vanquishing non-native plants and sometimes successfully growing natives in their place, and being a human multitool (from car and HVAC maintenance and construction to photography).

• Blitzkrieg Learning: Understanding what I don't know and need to know at super speed.
• Zero fear: No hesistation...ever.
• Asking questions no one else imagines: I just think differently.
• Seeking nuance: I'm suspicious of problems that look simple, because usually an important detail or consequence wasn't considered. Sometimes I'm pleasantly surprised when a solution easily falls into place.
• Breathing life into ideas: I'm crazy good at navigating ambiguous problems and chipping away at abstract ideas.
• Writing for humans: I'm obsessed with plain, approachable writing.

... all superpowers have a cost. You can imagine for yourself how each of those can backfire sometimes.

ChatHoard is a FOSS webapp that started as a tool to store and search my LLM chats (e.g. ChatGPT, Claude), but is rapidly growing in scope into a private, collaborative space to mine, organize, share and collaborate on LLM generated content... because I think I'm not the only one who needs a model-agnostic space to make the most of their data. It is built on FastAPI/Pydantic, React/ChakraUI, and uses Qdrant/DocArray for intelligent vector search. It's designed to be both locally deployable and also massively scalable.

I have a few projects I intend to get to a stable place eventually...

• A tool solving the problem of passwordless Linux logins
• Securely handles unlocking Gnome/KDE keyrings and wallets without a password. (Freedesktop.org SecretService is the underlying cryptographic management service -- it's dated... maybe some day I'll write a replacement.)
• Automates keyring unlocking using TPM 2.0 or Yubikey as a root key generator (SecretService only accepts passwords for keyrings/wallets, so AutoKeyring derives a secure hash to use as a password or derives a secret to encrypt and store a password for unlocking operations.
• Status: mid-refactor. I originally planned to use Clevis, but thought the implementation was hacky, so I'm refactoring as all-python.

• Mostly a tongue-in-cheek project to gain a deep practical understanding of AWS serverless app dev/deployment.
• Self-deploying AWS serverless system of microservices that simply checks amazon.jobs and provides better search settings and an alerts agent.
• The project emerged from my appreciation for irony because amazon.jobs doesn't offer a job alerts service and I thought...🤔 where better to build an amazon job alerts service than AWS?
• Status: Basic service works. I just need to finish the search template generation UI and email templating. Had I known pydantic existed it would have went much faster...
• Maybe one day someone at Amazon will happen upon it and use some of the library to implement an actual alerts service (they would only need a small portion because they have amazon.jobs already)...

• A pandas namespace extension (like .dt for Series) that adds Federal timeseries information to native pandas objects.
• The offsets module is finished and serves some niche calendar offset capabilties (Federal holidays accounting for Presidential proclamations, military leave passes, civilian and military paydays, federal fiscal years/quarters) that impact federal productivity, and actually adds vectorized operations where there weren't vectorized operations in pandas Offsets (another backburner is proposing those changes to pandas itself).
• There is also an accompanying database of funding/appropriations status for all top-level federal departments since 1970 (i.e. shutdowns, continuing resolutions, etc). The library aims to enable sophisticated statistical studies of the effects of federal budget and productivity processes on both government function and the unimagined consequences (i.e. impacts on local businesses, etc). I'm still thinking through how best to deliver the budget data for clean integration with pandas. Another path is taking it back a step, using python (or numpy) objects and creating an interface so it can integrate with not just pandas but polars and other dataframe libraries. My offsets work all built from numpy, so the latter step would be the easier of the two.