wballard

This is pretty cool - very close to https://shipula.userbugreport.com/ that I've been cooking up to make my own life easier.

Besides reduction to search — solr / elastic / Lucerne / xapian, which is the most common approach I have used commercially, my actual favorite is precomputation.

At the moment, keras embedding model, multiprocessing, annoy, and emitting csv (object id, other object id, score) as a batch process and loading it in my database. Queryti recommend. This trades a prebuilt for near instant runtime and — near Nothing net new to break.

I’m working at commercial — 2-5 million item — scale, not ‘internet scale’ billions of items.

Hope that helps.

We’ve been using our own setup for 4 years now. https://github.com/wballard/starphleet

We have 200 services, counting beta and live test variants. Most of the difficulties vanished once we had declarative versioned control of our service config in the ‘headquarters’ repository.

Not aware of anyone else using this approach.

That is the best comment I’ve seen all year. Bravo.

Hmm — fake analogy. Bridge fall down would be more on the scale of, a software install stopped all service permanently, self erased, and could never be restarted.

Real software quality issues are more like re paving, painting, new cables, scraping off rust, closing for inspection — and slow commute traffic over the bridge.

Argument by analogy is just sophistry, the real argument is to precisely account for the total economic cost of defects, and what the cost would be to correct them. But that is a lot more effort than I have ever seen in a blog post.

Control software for the baby oil and baby powder plant.

Same here. Totally agree. Check out the new weldment and pipe features in OnShape, just noticed this last night.

Ahh sweet Rhetoric!

Arguments of sophistry and nationalism are not data, but bias. None deserve preference for the accident of birth. Over the whole course of the human experience, the idea of nation is a distinction sans difference.

What works for me having managed 20 years now:

-accept that different folks work at different speeds, measure success on dollars made or saved -- real business metrics -- not are they fast vs your expectations -share the end goal and business metrics, don't 'task' people, give the a mental stake in the real problem -- otherwise you are not getting their brains, just their hands -if you have a strong preference for hands off and not guiding, hire for that, which is a longer chat but I can share how I do it -if you are willing to have 'dependent' developers, personally pair with them and see the real problem in real time, not a weekly or daily vignette

This is a cool paper, I love recommeender ideas. I've always set up a multi algorithm bake off for live systems. And every time, showing a block of popular links, either globally, or in the category gathers more clicks than any smarter algorithm. YMMV, but I always found this depressing!

Popular is popular is clicks.

Interesting points, but I feel that the root cause of these process debates is one un-natural act: estimation. Personal opinion is that far too much energy is put into processes to generate predictability, which is only practical when you have done it before, repeatedly.

All this focus on estimation, points, timing -- it just doesn't create what is needed to both do a great job and go fast: clarity.

This is why I don't run my shop with deadlines. I have no experience that deadlines/sprints/iterations make things faster, or create better ideas -- the things I'm actually interested in.

Guys, rich people don't keep a Scroogey-McDuck pile, large concentrations of funds make venture possible, as it lowers the marginal risk. Try having our current startup fueled cycle without concentrated wealth.

Can we really drive the human experience and technology higher by diluting our funds in an effort to normalize and support, as taxation would do creating obs the average can fill -- or should we concentrate and invest in those with the most potential?

I hear you, but that's not in your own interest to feel that way. Honor in this case is meeting your commitments, your explicit commitments. In this case, you are -- at least in you words here -- negotiating against yourself. Count on the owner and your boss to act in their own interest, for themselves.

"Let me check my confidentiality agreement to make sure that won't get me in trouble and I'll get back to you"

Hmm, so a raise in no way implies a commitment to stay. Raise your sights -- going from 'broke' to 'getting by' may be motivating, but not quite as lucrative as going from 'getting by' to 'killing it'.

I'll tell you, flat out -- no matter how much you make, you'll feel the pinch, you'll just learn to move the 0.