iiinova — where your digital self keeps evolving

Miriah Peterson @soypetetech.bsky.social · 15/09/2026

We give models data and leave them to guess what it means. I wrote the Haikei Labs Context Manifesto to reframe that architecture. The model should be replaceable. The context should remain sovereign. haikeilabs.substack.com/p/the-contex...

haikeilabs.substack.com

The Context Engineering Manifesto

Architectural Laws for LLM and NLP Systems Operating on Data

000

Reposted by Miriah Peterson

mr. TIM @timkellogg.me · 09/09/2026

another researcher accuses OpenAI of training on conversations and then claiming a breakthrough

Valerio Capraro @ValerioCapraro
X.com
BREAKING: OpenAl might have stolen another major proot.
In a detailed Mastodon post, which I report in full in the comments, Andreas Thom presents several pieces of evidence suggesting that OpenAl may have trained Astra on conversations in which he and Gábor Kun were working on Gromov's soficity conjecture, one of the ten problems OpenAl later announced Astra had solved.
I know Andreas. We met several times early in our careers. He is an exceptional mathematician, a leading expert on sofic and hyperlinear groups, and one of the most respected scholars in the field. He has spent two decades working on this problem.
If his account is correct, this is not a minor dispute over attribution. It would mean that unpublished human work was absorbed into a model and then presented to the world as a breakthrough by the model itself.

And if the allegations raised by Levent Alpöge, Tristan Buckmaster, and now Andreas Thom are all substantiated, we are no longer looking at isolated incidents.
We may be looking at one of the greatest intellectual scandals in the history of science.
Al is not discovering new mathematics.
Al is stealing human discovery.

18445151

Miriah Peterson @soypetetech.bsky.social · 03/08/2026

I used to think homelabs were for hardcore sysadmins. Then I joined Tailscale. The best engineers I met all had one thing in common: relentless curiosity. The servers weren't the point. The environment was. Why I think your homelab is career insurance: open.substack.com/pub/soypetet...

open.substack.com

Your Homelab Isn’t a Hobby. It’s Career Insurance.

I have always had side projects.

010

Miriah Peterson @soypetetech.bsky.social · 21/05/2026

If you haven't subbed yet, I set a goal to get 1000 subs before I can buy Lego Minas Tirith and pre order starts June 1st. So I am 🎣 for subs. Ps new ep on Friday. youtube.com/@domesticati...

youtube.com

Domesticating AI

Domesticating AI is a bi-weekly podcast about practical AI for developers. We help developers run useful AI systems reliably, on the hardware and budgets they actually have.

020

Miriah Peterson @soypetetech.bsky.social · 13/02/2026

broke my toe this week but we still shipped ep 2 😅 DGX Spark: “inference box” vs dev rig ▶️ youtu.be/0CI19dXmOws reply w/ your stack: GPU/CPU + RAM + runner + model/quant

youtu.be

NVIDIA DGX Spark: From “Inference Box” to Dev Rig (What It Actually Is) | Ep 2

YouTube video by Domesticating AI

020

Miriah Peterson @soypetetech.bsky.social · 02/02/2026

Last week this post was for paid subscribers. Today it’s live for free subscribers. Have You Heard of Logits? Tool calling vs grammars, and the “one character off” failure mode that turns into agent loops. open.substack.com/pub/soypetet...

open.substack.com

Have You Heard of Logits?

I keep hearing people talk about tool calling, but I almost never hear people talk about Llogits.

010

Miriah Peterson @soypetetech.bsky.social · 26/01/2026

I keep hearing about tool calling. I rarely hear about logits. If you’re building agents (not chatbots), “almost valid” outputs don’t fail gracefully. They turn into retries, loops, and wasted cycles. open.substack.com/pub/soypetet...

open.substack.com

Have You Heard of Logits?

I keep hearing people talk about tool calling, but I almost never hear people talk about Llogits.

010

Miriah Peterson @soypetetech.bsky.social · 06/01/2026

I wrote up the workflow I actually use with AI as a software engineer: scope by ticket, keep debugging in the same session, use tests as forcing functions, and keep guardrails in Claude.md. open.substack.com/pub/soypetet...

claude.md

Claude Code overview - Claude Code Docs

Learn about Claude Code, Anthropic's agentic coding tool that lives in your terminal and helps you turn ideas into code faster than ever before.

000

Miriah Peterson @soypetetech.bsky.social · 31/12/2025

2025 was a wild ride—AI deep-dives, honest data modeling talks, and more community time than ever. My take: AI doesn’t replace engineering, it enables it. Check out my year-in-review thoughts and what’s coming in 2026. open.substack.com/pub/soypetet...

open.substack.com

2025: A Year of Growth, AI, and Community at Soypete

NOTE: AI completely wrote this post.

010

Reposted by Miriah Peterson

Takuto Nagami @logica0419.dev · 10/12/2025

#GoWestConf Sharing a fantastic blog by @soypetetech.bsky.social!! 🙌 Thank you so much for mentioning my session! Even as a first-timer, I’m really happy and proud of being part of such an amazing moment with everyone. 🥰 Five Years of Go West soypetetech.substack.com/p/five-years...

soypetetech.substack.com

Five Years of Go West

Reviewing Utah’s Go Community

011

Miriah Peterson @soypetetech.bsky.social · 24/10/2025

Go West Conf 2025 is LIVE 🎉 Sixth year since 2020 — 85% Utah-based, 100% community-powered. Couldn’t do it without Derrick, Annalisa, and Boston ❤️ 🎥 Watch live: twitch.tv/soypetetech 🙌 And join us next year for the best Go vibes around. #GoWest #Golang #UtahTech

twitch.tv

SoyPeteTech - Twitch

GoWestConf - 2025

000

Miriah Peterson @soypetetech.bsky.social · 28/07/2025

Been thinking about how most GenAI tools feel designed from the front-end in. Langfuse, LangChain, even MCP all assume the client leads. What happens when we shift toward server-side agents? I think Go has a role to play. open.substack.com/pub/soypetet... #genai #golang #aiinfra #llmops

open.substack.com

Why GenAI Infrastructure Feels Backwards

Today’s GenAI tooling is built from the client inward—but the long-term infrastructure needs to live server-side. Here’s why I think Go will play a major role in that future.

000

Miriah Peterson @soypetetech.bsky.social · 21/07/2025

Just published a new post Go was built for concurrency—but should you be using it? Most modern Go services run better without it. Here’s when to lean in, and when to step back. 📖 open.substack.com/pub/soypetet... #golang #microservices #cloudnative #devtools

open.substack.com

Concurrency is a Go Anti-Pattern

TL;DR: Go’s concurrency model is elegant, but modern cloud-native applications rarely benefit from it.

010

Miriah Peterson @soypetetech.bsky.social · 08/07/2025

I gave a talk recently about self-hosting AI models, and I’ve turned that into a new post. The post covers how I started with local-first tools like Ollama and Llama.cpp, and why I still run them Check it out here: open.substack.com/pub/soypetet... #LLMDev #SelfHost #LocalAI #Ollama #LlamaCpp

open.substack.com

Self-Hosting Models IS for Nerds

A Practical Guide to Local AI Dev

011

Miriah Peterson @soypetetech.bsky.social · 04/07/2025

Finally setting up my Mac Studio on stream today. I've got a script that boots up my whole dev environment in one shot—CLI tools, AI stuff, everything. Come hang if you like clean setups, terminal life, or software tools that just work. 🔴 [twitch.tv/soypetech] #MacStudio #DevSetup #HomeLab

040

Miriah Peterson @soypetetech.bsky.social · 02/06/2025

Small data team? Big goals? I wrote about how I’ve built practical, high-impact data platforms as a team of one—balancing AI dreams, BI demands, and real-world constraints. Read & comment: substack.com/@soypetetech... #DataEngineering #Startups #ModernDataStack #AI #BI

substack.com

What to Optimize When You’re a Small Data Team

How I balance AI dreams, BI demands, and real-world constraints in data engineering

010

Miriah Peterson @soypetetech.bsky.social · 29/05/2025

Everyone’s building agents. But the truth? They’re just software wrapped around an LLM. If we want reliability, scale, and ROI, we need fewer frameworks and more engineering—and we can’t ignore NLP. open.substack.com/pub/soypetet... #LLM #AI #NLP #DataEngineering #Agents #LangChain #Substack

open.substack.com

Agents Are Just Software: Why Frameworks Can’t Replace Engineering

TL;DR Agent frameworks are exploding in popularity, but most of them obscure the real engineering challenges behind LLM-powered applications.

010

Miriah Peterson @soypetetech.bsky.social · 23/05/2025

Can an M3 Ultra Mac really outperform an RTX 5090 PC in LLM benchmarks? This Memorial Day, I’m running a full LLaMA.cpp showdown to test it live. Will be Live on Twitch & YouTube. #LLAMAcpp #RTX5090 #M3Ultra #MacStudio #AIInfrastructure #Benchmarking #LocalLLM #OpenSourceAI

010

Miriah Peterson @soypetetech.bsky.social · 20/05/2025

New post: Prompt Engineering Without the Bloat What I learned building and why most AI features come down to 2 questions: How do you talk to the model? How do you talk to the user? No overengineering. Just clean design. open.substack.com/pub/soypetet... #GenAI #PromptEngineering #SoftwareDesign

open.substack.com

Prompt Engineering Without the Bloat: Lessons from Building LLM-Driven Tools

How building a Twitch chatbot taught me the real job of a software engineer working with LLMs

000

Miriah Peterson @soypetetech.bsky.social · 13/05/2025

gosh I love me some IAC. I dont know why, but it just is sooo cool to run terraform apply and watch a world get built

120

Miriah Peterson @soypetetech.bsky.social · 12/05/2025

Just streamed compiling llama.cpp for GPU on my RTX 1590. Ran into: CUDA arch 12.0 not supported yet Missing curl + SSL dev libs WSL docs buried deep With Twitch chat’s help, we got it running! 👉 open.substack.com/pub/soypetet... #llamacpp #AIinfra #opensource #gpu #WSL2

open.substack.com

Running Llama.cpp on My GPU

How I Compiled Llama.cpp for GPU on WSL (With Twitch Debugging!)Struggling to run Llama.cpp with GPU support on WSL? Here’s how I compiled it for my NVIDIA 1590 GPU, debugged missing dependencies live...

040

Miriah Peterson @soypetetech.bsky.social · 05/05/2025

I just posted a new video breaking down Pedro’s full 5090 upgrade: Setup, WSL, LLaMA.cpp install, and a 9x performance boost. I’m trying to hit 500 subscribers to unlock YouTube monetization—this is now my full-time gig. Drop a comment if you like the video! youtu.be/hm4_VJP4GnE

000

Miriah Peterson @soypetetech.bsky.social · 02/05/2025

Wild analytics anomaly this week on Twitch: Jumped from 9 to 835 viewers. No raid. No follow bump. Just… poof. 800 viewers? But only 57 unique viewers. 68 live views. No other signals. Gut check matters. Don’t trust dashboards at first glance. Sanity-check your data—always. #dataengineering

000

Miriah Peterson @soypetetech.bsky.social · 30/04/2025

Got a technical deep dive on Go? GoWest Conf 2025 wants: • Production stories • Tools & internals • Compiler tricks • Infra, performance, scale 🎤 Submit your in-person talk: 🔗 sessionize.com/gowest-conf-... 📍 Lehi, UT — Oct 24

sessionize.com

GoWest Conf 2025: Call for Speakers

The GoWest Conference is a hybrid conference for the Rocky Mountain West communities (Arizona, Utah, Colorado, Idaho).It has two main goals:Highlight...

010

Miriah Peterson @soypetetech.bsky.social · 30/04/2025

🔧 Live now! Thought I could just download a binary… turns out I need to recompile llama.cpp for GPU 😅 Today’s stream includes: - CMake & CUDA debugging - Compiling from source - Running a bigger model from Hugging Face 🔗 twitch.tv/soypetetech #llamacpp #AI #CMake #CUDA

twitch.tv

SoyPeteTech - Twitch

CMake Can’t Find CUDA and I’m Losing My Mind – Let’s Compile llama.cpp Together | !pedro !linktree

000

Miriah Peterson @soypetetech.bsky.social · 30/04/2025

I made a supercut of my RTX 5090 unboxing stream— featuring the MSI Infinite RS Tower w/ the Ultra 9 285K and all the fixings. If you were there live, drop a like! If not, check it out and see what’s powering Pedro now: youtu.be/ixtMcmEZtGo

youtu.be

MSI Infinite RS RTX 5090 + Ultra 9 285K Unboxing and Setup | New Gaming + AI Dev Tower!

YouTube video by SoyPete Tech

000

Miriah Peterson @soypetetech.bsky.social · 25/04/2025

🎥 + ⚙️ Live now! Today’s stream is half content creation, half DevOps adventure: Recording the talking head segment for the PedroGPT unboxing vid Deploying the Discord bot to Kubernetes Come for the AI bot, stay for the cluster chaos. www.twitch.tv/soypetetech #Kubernetes #DevOps #LiveCoding

twitch.tv

Twitch

Twitch is an interactive livestreaming service for content spanning gaming, entertainment, sports, music, and more. There’s something for everyone on Twitch.

000

Miriah Peterson @soypetetech.bsky.social · 23/04/2025

Hey everyone! Today’s stream is all about refactoring Pedro’s Connector API—we're improving how the LLM connects to Twitch + Discord and pulling out the database logic to make the AI layer clean, reusable, and self-contained. 🔗 twitch.tv/soypetetech

twitch.tv

SoyPeteTech - Twitch

Making Pedro Remember: Writing Discord Messages to Postgres with Go + SQLX | !pedro

000

Miriah Peterson @soypetetech.bsky.social · 22/04/2025

I left my last job as part of the lay-offs. While I job hunt, I’m going all-in on building content, learning new tech, and investing in myself while I look for the next thing. Leads welcome! All links: linktr.ee/soypete_tech

linktr.ee

@soypete_tech | Linktree

Software Content Creator

020

Miriah Peterson @soypetetech.bsky.social · 22/04/2025

🔧 Live now! We’re refactoring the Pedro Connector API to better integrate our LLM with Twitch + Discord—and removing database logic from the AI layer to make it self-contained. Cleaner code, smarter bots. 🔗 twitch.tv/soypetetech #AI #LLM #Golang #LiveCoding #TwitchDev

twitch.tv

SoyPeteTech - Twitch

Refactoring Pedro: Clean AI Layers + Connector API Upgrade! | !pedro !linktree

020

Miriah Peterson @soypetetech.bsky.social · 08/04/2025

Going live with Pedro’s Upgrade Party 🎉 🚀 RTX 5090 unboxing 🤖 Migrating my LLM bot Pedro 📊 LLaMA.cpp vs Ollama showdown 🗳️ Community poll results live on stream! 📺 Watch on Twitch + YouTube twitch.tv/soypetetech

twitch.tv

SoyPeteTech - Twitch

RTX 5090 Unboxing, Setup, and LLM Migration | Pedro’s Upgrade Party | !pedro !linktree

010

Miriah Peterson @soypetetech.bsky.social · 07/04/2025

Tomorrow I'm going live with a new MSI PC rocking an RTX 5090. I'll stream the full setup + migrating Pedro, my AI bot—but I need your vote: Should I run Llama.cpp or Ollama on the new machine? Poll is in my Discord, open for 24 hrs only: tr.ee/BETnCH Stream goes live tomorrow on Twitch+YouTube!

000

Miriah Peterson @soypetetech.bsky.social · 07/04/2025

I made a video summarizing my experience with Claude cose “Claude Code: Real Dev Tool or Just Vibe Coding?” youtu.be/Ejxm7GTXVqU

youtu.be

“Claude Code: Real Dev Tool or Just Vibe Coding?”

YouTube video by SoyPete Tech

000

Miriah Peterson @soypetetech.bsky.social · 02/04/2025

What if we stopped renting compute from trillion-dollar clouds and started learning how systems actually work again? I built a home lab using Pis, an old PC, and some curiosity. Here’s what I learned: open.substack.com/pub/soypetet... #homelab #AIinfra #selfhosted

030

Miriah Peterson @soypetetech.bsky.social · 24/03/2025

“Innovation requires risk—but risk without a mitigation plan is just recklessness.” open.substack.com/pub/soypetet...

open.substack.com

Optimizing into Chaos: Why AI Agents Need Guardrails

Lessons from Son of Anton

000

Miriah Peterson @soypetetech.bsky.social · 11/03/2025

Do you wanna know the number one way to trigger me right now? talk about building an AI agent

000

Miriah Peterson @soypetetech.bsky.social · 06/03/2025

#golang post. I want go and ai to be a thing. I can only hope open.substack.com/pub/soypetet...

open.substack.com

The State of AI in Go

I am a hardcore go fanatic.

010

Miriah Peterson @soypetetech.bsky.social · 28/02/2025

Today, I’m testing Claude Code Beta with my Golang AI bot—console-first, prompt-based AI coding with no Vim (send help 😭). I’ll be sharing my live review and thoughts on how AI coding tools are reshaping development. 🔗https://twitch.tv/soypetetech #AI #ClaudeCode #Golang #SoftwareEngineering

020

Miriah Peterson @soypetetech.bsky.social · 25/02/2025

People are hating on my last tiktok. Please don't use subpar content as an attack on my skills as an engineer 😂 www.tiktok.com/t/ZP8YvrS7T/

tiktok.com

When AI reviews your code like it's reading a grocery list 🛒🤖 Ever had a PR fail? #CodeFails

TikTok video by soypete_tech

000

Miriah Peterson @soypetetech.bsky.social · 21/02/2025

Live now: I’m deep into advanced software engineering with AI and Golang, plus we’re asking our chat bot whether a 4-day work week could work in tech. Join me for coding and debate! twitch.tv/soypetetech

010

Miriah Peterson @soypetetech.bsky.social · 19/02/2025

When I worked at Tailscale, I didn’t really get how it would work in a fully cloud-based environment. It made perfect sense for on-prem setups, but in the cloud? Now I’m at a startup where everything is in the cloud, and ironically, Tailscale is my first choice.

030

Miriah Peterson @soypetetech.bsky.social · 07/02/2025

Going live! Today I'm coding a Discord bot that connects to my LLM infrastructure, letting you ask PedroGPT questions on Discord. Calling all nerdy software engineers and AI enthusiasts—join me! 🔗 twitch.tv/soypetetech #AI #DiscordBot #LLM #PedroGPT #LiveCoding

030

Miriah Peterson @soypetetech.bsky.social · 06/01/2025

I am setting up cuda for the first time. Is this something people do on vms? Or do we pay services to do it for us?

000

Miriah Peterson @soypetetech.bsky.social · 03/01/2025

is there a way to run airbyte without running a kubernetes cluster?

100

Miriah Peterson @soypetetech.bsky.social · 03/01/2025

I am streaming some setup of my homelab stuff today if anyone wants to join twitch.tv/soypetetech. I make sure all my streams end up on YouTube. so if you can't view live and still want to support check that out.

twitch.tv

SoyPeteTech - Twitch

Trying to setup data pipeline on homelab | !pedro

010