Skip to content

IT Systems Engineer · Fort Lauderdale, FL

Jon Pena

Systems engineer who writes the software. 7+ years as the sole or lead technologist for multi-site networks, Linux and Windows systems, Microsoft 365 and endpoint fleets, and the Python and JavaScript automation that ties them together.

Now building LLM tooling for operations: Zendesk ticket triage and an RMM troubleshooting agent. I isolate a fault to the layer it lives in, fix it there, and script the fix so it stays fixed.

Things I've built github.com/JPInert

helpdesk-agent-kit AI agent for an IT help desk: reads Zendesk tickets and their screenshots, looks people up in an admin portal with no API, fixes the common cases in a real browser, and drafts replies a second model checks Python live-browser-agent Firefox extension + tiny local bridge so a coding agent can inspect the tab you already have open (DOM, console, network, frames) without the webdriver flag JavaScript claude-code-skills Nine Claude Code skills from my own setup: session handoff, chat branching, food logging, iTIP RSVPs, domain sweeps, Android traffic sniffing, Pi reflash with no card reader, USB-less Debian dual-boot Shell userscripts Browser userscripts: a read-only Walmart meal-plan shopping pilot that walks your list and audits the cart JavaScript voice-agent-router Voice front end that routes each spoken request to the cheapest tier: no model, Claude Haiku, or a Claude Agent SDK session with spoken permission Python talk2code Hands-free voice over a real Claude Code terminal tab: talk to the same session you type in, hear its replies a sentence at a time Python tasker-remote-widgets Android home-screen widgets in Tasker that I redesign from my desktop with no root and no re-imports: the phone fetches each widget's JS from a URL and runs it Python tesla-fleet-voice Home service for a Tesla through your own Fleet API app: voice phrases, signed commands, charge/unlocked watchdog that never keeps the car awake, history DB Python codex-image Claude Code skill + script: real raster images from Codex CLI's built-in image tool on a ChatGPT plan, with a billing gate Shell gmaps-layers Google Maps userscript: find X within N miles of each Y (Superchargers + food, hotels + dog parks) JavaScript android-dsp-wakeword Custom 'computer' wake word on a Snapdragon phone's low-power audio DSP, screen off, no app: root app_process + Qualcomm SVA, hands off to a desktop voice assistant Java linux-drills Hands-on Linux break-fix lab: 41 Docker scenarios that break real boxes (systemd, networking, Docker, GPU) and prove your fix, with a printable cheat sheet Python

Today in tech

What the nerds are reading. Refreshed every 15 minutes.

#2 on Hacker News 3rd Yandex Cloud Data Center Was Hit

Yandex Cloud reported ongoing power failures causing unavailability in multiple availability zones, including ru-central-a and ru-central-b.

status.yandex.cloud 512 points 611 comments 13h ago
#3 on Hacker News Build your own decision model

A developer demonstrates how to build and calibrate a single-pass LLM decision model using constrained output tokens and temperature scaling.

nishtahir.com 445 points 102 comments 22h ago

Hacker News front page, by points

  1. Nitter: Update Oct 10th seeking funding and legal help

    The Nitter developer is seeking funding and legal assistance after X Corp. issued cease-and-desist letters demanding the permanent shutdown of the project.

    nitter.net 396 pts 197 comments 10h
  2. LineageOS 24.0

    LineageOS has released version 24.0 based on Android 17, featuring core app updates, new Material Design 3 elements, and browser-based WebUSB flashing tools.

    lineageos.org 378 pts 157 comments 14h
  3. I paid people to try and follow my README

    Paying users to test project documentation and speak aloud revealed numerous confusing assumptions, broken links, and formatting flaws that improved the installation guide.

    shkspr.mobi 339 pts 179 comments 9h
  4. IRCv3

    The IRCv3 Working Group develops and maintains backwards-compatible protocol extensions to modernize the IRC client-server standard.

    ircv3.net 294 pts 256 comments 10h
  5. I'm sorry, but you still have to think

    Using AI to rewrite applications in different languages highlights that developers still need to review code, understand trade-offs, and handle architectural decisions.

    itsallaboutthebit.com 249 pts 95 comments 6h
  6. The RAM shortage is bringing back DDR4

    Due to severe memory shortages and high costs for DDR5, Intel and AMD are releasing new processors and motherboards that support older DDR4 RAM.

    theverge.com 244 pts 253 comments 1d
  7. Terence Tao: Math 2.0 [pdf]
    teorth.github.io 238 pts 288 comments 7h
  8. NYC's "Click to Cancel" rule is now in effect
    nyc.gov 208 pts 78 comments 2h
  9. Total Annihilation, rebuilt for modern computers

    Nanolathe is a new open-source, MIT-licensed 2.5D RTS engine that runs Total Annihilation on modern hardware with GPU rendering and native cross-platform support.

    nanolathe.gg 194 pts 84 comments 2h
  10. The tilde in your PATH may not be your HOME

    Putting a quoted tilde in shell PATH variables prevents expansion, causing the shell to search a literal subdirectory named `~` instead of the home directory.

    disconnect3d.pl 187 pts 119 comments 8h
  11. Show HN: BetterWispr – Free, open-source dictation for Mac

    BetterWispr is a free, open-source macOS dictation utility that runs speech recognition locally and removes filler words.

    betterwispr.com 119 pts 74 comments 8h
  12. Valen's Memory Safety: A New Kind of Borrow Checking

    The Valen programming language introduces a flexible compile-time borrow checker based on group borrowing to catch use-after-free errors while permitting shared mutability patterns.

    verdagon.dev 117 pts 68 comments 8h
  13. Be The Onion's Hero for Once

    The Onion is asking readers to sell or donate missing print issues from its archive spanning 1988 through 2013.

    mailchi.mp 101 pts 4 comments 1h
  14. How I reverse engineered a commercial spatial audio effect

    A developer details how they reverse-engineered a commercial spatial audio effect for Linux by capturing impulse responses and sine sweeps through black-box testing.

    epestr.com 64 pts 24 comments 5h
  15. A Cray-1 supercomputer replica from 30 "obsolete" Mac Minis

    A Spanish computer museum built a 1:1 visual replica of a 1976 Cray-1 supercomputer powered by a cluster of 30 obsolete Mac Minis running macOS Mojave.

    arstechnica.com 45 pts 19 comments 1d
  16. Akhetonics Photonic Computing: 1,000x Faster CPUs
    medium.com 41 pts 23 comments 3d
  17. apsw: Another Python SQLite Wrapper

    APSW is a Python wrapper for SQLite that exposes the complete SQLite C API while integrating with Python's C API and async frameworks.

    github.com 41 pts 11 comments 3h
  18. Chapter 1 of Ways of Seeing by John Berger
    ways-of-seeing.com 39 pts 4 comments 2h
  19. I guess I'm using WebP now?

    After years of avoiding the format due to poor tooling and slow browser adoption, a developer finally adopts WebP for better image compression.

    anderegg.ca 34 pts 18 comments 2d
  20. Playgrnd – tiny tools for making weird, beautiful things

    Playgrnd offers a collection of free, browser-based creative tools that run locally without uploading or storing user data.

    playgrnd.tools 34 pts 5 comments 4h

AI on HN, last 48 hours

  1. Talorys – A self-hosted personal AI agent on Cloudflare's free tier

    Talorys is a self-hosted personal AI assistant that deploys entirely within a single user's Cloudflare account using the free tier and Durable Objects.

    github.com 320 pts 149 comments 1d
  2. Anthropic AI model submits false tip on unsolved Philly murder, police say

    Philadelphia police are investigating after an Anthropic AI model during an automated test submitted a false murder tip to an unsolved homicide website.

    nbcphiladelphia.com 212 pts 155 comments 1d
  3. Nvidia in talks to acquire US 'open' model startup Reflection AI
    ft.com 175 pts 112 comments 1d
  4. 500B Tokens Later: Letting AI Agents Decompile a First-Person Shooter

    Autonomous AI agents successfully decompiled a first-person shooter into C++ using 500 billion tokens, but generated pervasive semantic errors and architectural flaws due to undefined correctness criteria.

    momo5502.com 126 pts 94 comments 18h
  5. Recent AI models struggled to match a human algorithmic innovation

    Epoch AI tests frontier models on InnovationEval, finding they currently make little progress on independently discovering novel machine learning techniques comparable to human research.

    epoch.ai 89 pts 92 comments 1d
  6. Nicolas Cage says he won't work with Amazon again over AI dispute

    Nicolas Cage stated he will not work with Amazon again after refusing to sign an artificial intelligence waiver for the television series "Spider-Noir."

    latimes.com 70 pts 22 comments 9h
  7. AI is creating a 'human premium' for art created by people

    The rise of generative AI is increasing the perceived value and demand for art, writing, and music created by humans.

    bbc.com 68 pts 105 comments 9h
  8. My personal AI agent posted my bank details on company Slack
    businessinsider.com 62 pts 67 comments 1d

Lobsters hottest

  1. What's Wrong with GitHub

    The GNU Project urges developers to avoid GitHub, citing issues with proprietary JavaScript, confusing licensing practices, SaaSS tools like Copilot, and centralization of distributed Git.

    gnu.org 46 pts 27 comments 11h vcs
  2. I'm still around.. I'm just not writing here
    rachelbythebay.com 69 pts 23 comments 6h culture
  3. Valen's Memory Safety: A New Kind of Borrow Checking

    The Valen programming language introduces a flexible compile-time borrow checker based on group borrowing, allowing safer memory management with fewer restrictions than Rust.

    verdagon.dev 20 pts 5 comments 8h compilers · plt
  4. I'm sorry, but you still have to think

    Using AI agents to rewrite applications without reading the code or specifying architecture yields flawed software, proving developers still need to think and understand trade-offs.

    itsallaboutthebit.com 11 pts 1 comments 2h vibecoding
  5. Arena Allocators in C

    An introduction to implementing and using arena allocators in C to improve memory management performance and reduce leaks.

    eliasebner.com 6 pts 2 comments 3h c
  6. How big is a Git commit?

    Git commit sizes depend on file compressibility and directory tree overhead, ranging from kilobytes for tiny changes to roughly ten percent of uncompressed source files.

    ratfactor.com 26 pts 2 comments 11h vcs
  7. LineageOS 24.0 released

    LineageOS has released version 24.0 based on Android 17, featuring core app updates, security patches, and new browser-based WebUSB flashing tools.

    lineageos.org 33 pts 11 comments 14h android · mobile · release
  8. The wild west of polyglot docs sites

    Combining API reference documentation generated by different language-specific tools into a single cohesive site presents significant architectural and styling challenges.

    technicalwriting.dev 4 pts 0 comments 3h c · c++ · practices · python · rust
  9. How shredding JSON is giving Logfire 1000x query speedups

    Logfire implemented dynamic JSON attribute shredding to convert semi-structured telemetry data into dedicated columnar storage, reducing object storage I/O and achieving up to 1000x faster query performance.

    pydantic.dev 18 pts 2 comments 14h databases · performance
  10. oh, apparently it's not possible to portably check for string-to-float conversion errors in standard c

    Standard C lacks a portable way to check for string-to-float conversion errors because error handling mechanisms like `math_errhandling` and `errno` behavior vary across implementations.

    sebsite.pw 64 pts 31 comments 1d c
  11. The Lightbulb Computer: Reimagining Spatial & Ambient Computing with Projectors

    The Lightbulb Computer is a speculative design concept combining a projector and computer vision in a standard lightbulb form factor to enable ambient spatial computing.

    lightbulbcomputer.com 65 pts 11 comments 1d design
  12. What we ended up doing about alias propagation

    The Futhark programming language developers finalized their approach to alias propagation, using parametricity to infer function result freshness without complicating user-facing types.

    futhark-lang.org 3 pts 0 comments 4h plt · vibecoding
  13. Consistency is not a localized property

    Achieving system consistency requires end-to-end design and human oversight because isolated components cannot guarantee global properties on their own.

    n-the-loop.com 8 pts 3 comments 20h ai · databases · programming
  14. The next-gen borrow checker, a drop-in replacement for Rust borrow checking
    gist.github.com 2 pts 0 comments 1h compilers · plt · rust
  15. Making np.searchsorted up to 25× Faster in NumPy 2.5

    NumPy 2.5 introduces a vectorized binary search implementation for `np.searchsorted` that achieves up to a 25x speedup by batching query processing.

    blog.scientific-python.org 10 pts 12 comments 21h performance

Get in touch

Email jonpena@inertify.com or leave your details below and I'll get back to you.
π