Self-directed systems & AI engineer, available for contract (or the right full-time role). In the last year — solo, mostly from scratch — I've built: an LLM training engine and active research project (AGPT); an AI agent framework that autonomously runs my own production infrastructure (copious); a semantic-memory / RAG library (memo); a C11 YAML 1.2 parser that passes all 308 test-suite cases and outperforms every other compliant parser on real-world benchmarks (YAM); and a serialization format about half the size of JSON that parses ~3x faster (C0DATA). I also maintain Ruby Facets (since 2005). I go from model internals to production ops, and I ship — every link on the portfolio is real, running code. Research is my first love — AGPT above is my own line of it — and the work I'd most gladly relocate for, though I'm every bit as happy shipping production systems. The site also has an AI assistant you can talk to; she'll take a brief and send it to me.
I think it had more to do with the reality of the market. Solar panels have become incredibly cheap and that's because they are mass produced and standardized. Everything in the manufacturing process has been optimized. Now it is technically of course possible to make them other form factors, but artisinal solar panels are simply so much more expensive and cannot compete in any meaningful way with regular panels.
It was a bad idea from the beginning, technically and economically it sucked, the only possible utility was areas with strict heritage constraints which forbade normal rooftop PV.
What a ridiculous statement. People get all sorts of roofs with all sorts of trade offs. Looks is absolutely one of them. Also, any product that gets popular eventually gets highly optimized manufacturing pipelines bringing the cost down.
It doesn't need to compete with normal panels on only one metric. People will accept longer payoff times for aesthetics or durability if the ratio is right. Also, who really cares about Tesla at this point? Other companies are now producing these panels.
Like everything Musk, it died because of his poor business practices and his politics. The only thing he seems to excel at these days is extracting government money.
it's not really possible to stay human readable and get the compression levels and random access properties I was going for. But it is as human tooling friendly as possible given the constraints.
I find it obvious that your first attempt failed. Try again, you have not even remotely failed enough if you are making the argument that this is kinda readable. Yes, ascii words are easy to pick out, you didn’t do that, you did the part that makes it all harder.
CLI is an API, so I decided to treat it as such and created Jargon (for Crystal). Define CLI in JSON Schema, get argument parsing, validation, help text, and shell completions for free. Supports nested options with dot notation, subcommands, config file loading and merging, env vars, XDG base dir compliant, and JSON to stdin.
Honestly if it were not for my extensive Ruby background that I have now been able to carry over to Crystal, I probably would have dived into Guile.
(I have been enjoying Elixir too, but at the end of the day it doesn’t quite sit right with me — just feels a bit clunky. Gleam seems an attractive alternative though. The BEAM rocks, but it is a heavy dependency that doesn’t fit all distribution needs.)
Remote: Yes — or onsite/hybrid, any is fine
Willing to relocate: For the right role — especially research
Technologies: Rust, C, Crystal, Elixir, Ruby, Python, TypeScript; LLM training & agents, RAG / vector memory, high-performance parsers & data formats, full-stack + infra/ops
Résumé / Portfolio: https://tabcomputing.com/portfolio
Email: [email protected]
Self-directed systems & AI engineer, available for contract (or the right full-time role). In the last year — solo, mostly from scratch — I've built: an LLM training engine and active research project (AGPT); an AI agent framework that autonomously runs my own production infrastructure (copious); a semantic-memory / RAG library (memo); a C11 YAML 1.2 parser that passes all 308 test-suite cases and outperforms every other compliant parser on real-world benchmarks (YAM); and a serialization format about half the size of JSON that parses ~3x faster (C0DATA). I also maintain Ruby Facets (since 2005). I go from model internals to production ops, and I ship — every link on the portfolio is real, running code. Research is my first love — AGPT above is my own line of it — and the work I'd most gladly relocate for, though I'm every bit as happy shipping production systems. The site also has an AI assistant you can talk to; she'll take a brief and send it to me.