Is this self hosted or giving the cloud the keys to the kingdom?

Not really sure how to answer this, because there are varying degrees of "self hosted" vs. "cloud hosted".

This is a Next.js app hosted on Render.com, which is a managed VPS offering similar to Heroku. BudgetSheet is also of course completely reliant on Google Cloud though with Google Apps Script and the Workspace Marketplace where it is listed.

This makes sense if you think models will continue their current architecture and will always generate random predictions full of hallucinations. I see the next leap forward in SOA models with someone finding the right way to architect a Logic engine and a Knowledge engine necessary to implement the Provability Fabric you propose. In short, I would say a knowledge kind of bloom filter should weed out factual hallucinations, while a logic one will solve the "how many Rs in strawberry" problem.

Most likely a SOA model will implement a kind of provability fabric to set up guaranteed output rails. But if we implement this today, for example, let's say we tell a model to classify input into one of 3 set categories, the PF will guarantee the model will only output one of those three options, but it can still "hallucinate" by picking a bad category using faulty logic, breaking the rules, etc.

I don't mean to be rude, but I don't get how this is any better. Feels too manual to type "uv -add dep script.py" instead, I feel the automation tool I'm waiting for will scan my script, auto-import all the deps I'm calling in the script while ignoring the ones that I forget to use, AND set up the env with all the deps AND run the code in the same one liner. To me, uv add X is no different than running env/pip install requirements.txt.

Compare the before vs after

Before (analogous to go mod init):

    python -m venv venv
    source venv/bin/activate
    python -m pip install -U pip
    pip install httpx
    pip freeze > requirements.txt
    nvim foo.py
    # find a way to share foo.py and requirements.txt

On another machine (still the before scenario, this time analogous to maybe go run):

    python -m venv venv
    source venv/bin/activate
    python -m pip install -U pip
    pip install -r requirements.txt
    python foo.py

In the after scenario:

    uv run foo.py

That's it. Comparable to

    ./my-go-binary

What people like about this workflow is that you're not maintaining a separate venv or a separate requirement and it's declarative rather than imperative, this gives you two big advantages:

First, you can move that script to a different machine and do `uv run {script}`, no need to recreate a venv or provide install instructions (I believe uv will now even grab an appropriate version of Python if you don't have it?). This comes from PEP 723, and multiple tools support doing this, such as hatch.

Second, when you "add" a requirement instead of "install" a requirement it manages that with the knowledge of all requirements that were added before. For example, if I `pip install foo` and then `pip install bar` pip does not consider foo or it's dependencies as required when installing bar, so it's possible that you can break `foo` by installing completely incompatible dependencies. But when you "add foo" and then "add bar" from uv (and other tools that are declarative, like Poetry) your environment gets updated to take everything into account.

If managing Python dependencies is second nature to you then these might seem like extra concepts to keep in your head, but lots of people do find these useful because they find they can think less about Python dependencies.

Location: United States, US citizen

Remote: I don't mind in office or remote. Preferably hybrid or occassional meets in person. Willing to relocate: Yes, to another state in US, would consider other countries

Technologies: Python, TS/JS, MSSQL, PostgreSQL, MS PowerBI, PowerApps, Salesforce+Tableau, Metabase, AI work automation

Language: native English & Spanish speaker, ex-Interpreter (CHI certified)

Titles: Technical Operations Manager, Business Intelligence Analyst, Solutions Architect, Data Analyst, Operations Engineer, Technical Project Manager, IT Consultant

I'm an ex-Developer turnes Data Analyst consultant. I help businesses and public agencies improve their operations through industry-leading management analytics strategies. Looking for management or consulting type roles where my skillset can help technical teams operate better and help executive management take better decisions.

Email: Contactme@myhnusername.com

Thanks for sharing your experience I’m glad to hear you're finding value in understanding muscle targeting.

Bcs It's true that most beginners tend to think in terms of "full body" or "upper body" rather than doing a structured split let's say. They don’t usually say "I want to train my posterior deltoids and lats" lol

I love the idea of filtering exercises by gym type or gym but can be hard to handle for "private" gyms and will also need some kind of moderation... Could work for large branded gyms though.

This looks great. Regarding the taglines and copy being confusing for some, I understood it, but I have some suggestions:

Instead of saying what it is, say what it does (for the user). For example instead of 'Offline-first API Client'..., try "Document your API, test it, and version it using Git 100% locally and free in a modern GUI. Voiden even includes a terminal so you can execute any command or custom workflow without ever leaving the app. (OSS release coming soon)". Remember, show, dontt tell.

Much appreciate the suggestions! :)

My experience with PG ended a few years ago. The thing that killed PG for me is the lack of collation for searches. I haven't tested since the last updates that mentioned something along these lines, so this may have improved.

PG for years relied on unaccent which is not a true solution as it requires manually storing two versions of a text (the original and the un-accented one) and handle indexes + searches manually for each field... Not to mention unaccent itself is an extension you have to install. Whereas collation is built-in and it just works by handling it all under the hood in SQL Server. This feature alone is worth switching databases for my use cases.

That and the updates being a royal pain.

Thanks for sharing.

This is not about routers at all. I was expecting a comparison of models that run OpenWRT or similar.

What exactly is an LLM router and why would I use it? Most prices on this page are negative or 0 with tons of significant digits. How does that help?

OpenRouter is a service that lets you access many different LLMs (like OpenAI, Anthropic, etc.) through one interface. This site makes pricing comparison easier, which isn't straightforward on the OpenRouter site itself. The $0 price means that the model inference is currently free.

Thanks for the feedback, I've added filter out free models anyway.

That filter isn't entirely working for me - screenshot here: https://gist.github.com/simonw/f6614ef90f670818106f1d1a48212...

It's still showing models with $0.0000 as the price even after I selected "Hide free".

Ah, missed that one. It was only filtering models that contained 'free'. Fixed that in the current release

HR enthusiast entrepreneur here too. HR is a MASSIVE universe of fields such as Recruiting, Training, Policy & Procedures, Payroll, Staffing Needs Studies, Scheduling (future), Workforce Distribution (live scheduling), Workforce Management (KPIs from past performance), sometimes Roles & Permissions, not to mention being a system of record for critical business ops.

I'm sketching out a project that will cover many of these and I've developed similar projects before. If you need ideas, feel free to reach out. contactme at myuser.com

Disclaimer: I've never believed in your product or any similar tool neither as a dev, analyst or manager (10 YOE). I think they're useful for CS competitions, clubs and so forth, but overall I see it the same way people good at Excel or Chess win competitions but that doesn't make them good analysts/devs/employees.

IMO there is a huge gap in the market for facilitating in-person interviews that no tool has really exploited yet.

What would a solution like that look?