chrstphrhrt

Then the prototypes get handed off to the main engineers to run with, and support is provided. By the time the thing sees production usage, the original inventor has long since moved onto solving other problems on the horizon.

Praise is attributed to the last and loudest to have touched a project. Innovators are deeply satisfied by that and don't need the praise because it actually gets in the way.

https://status.github.com/messages

With web apps any telemetry data is pretty much limited to things on that domain. Browsers have access to more in principle.

I guess if the data transmitted is not personally-identifying or profile-building then it's fine. E.g. number of sites only, not urls. On second glance at the new FF page https://www.mozilla.org/en-US/privacy/firefox/ it seems like the stuff they collect is pretty innocuous. Paranoia ftw.

I had to read the privacy notice they open on start to find out that telemetry was enabled by default.

Hey folks, I'm one of the engineers at Finn. The product is a personal banking NLP + ML system with chat frontends. Our customers are tier one and two banks. We started on Facebook Messenger, later added Slack for human customer service, Google Assistant for voice, and most recently our own SDK for Android, iOS and web.

We're looking for new teammates on the product and data science teams.

On product (my team), we use and/or plan to use:

- node 7.10+ (tracking/testing latest stable) for multiple medium-sized, feature-oriented services

- postgres, redis, elasticsearch/logstash/kibana, low level drivers and query builders, no byzantine object mappers :)

- GraphQL

- ES2017+ without transpilers on the server

- Native functional/immutable style to avoid null and undefined issues

- react for web and react-native hybrid for mobile

- Experimenting with OCaml and ReasonML

On data science, we use and/or plan to use:

- python for services, tracking 3.5+ for type annotations and asyncio but mostly holding steady at 2.7 due to dependencies

- bottle + uwsgi for web, open to whatever

- Stanford NLP for dates and times

- NLTK for tokenizing/stemming etc

- Scikit for basic transformations

- pandas for tabular data

- numpy for matrices and the nice math syntax

- spark for the actual model

Current headcount for the company is ~42 people. We're from many different places, ethnicities, genders, and backgrounds. Women, LGBTQIA+, indigenous, racialized people, immigrants, people with disabilities and those from other marginalized groups are strongly encouraged to get in touch.

If you have shipped anything using any of the stuff above, send me an email to chris.hart@finn.ai to arrange a call.

Sometimes I wish there was a TrueHN or LongFormHN section that would separate the monkish/deeper stories from the mainline/fashionable links. For now bookmarks will do :)

I had chalked it up to some kind of national self-deprecation that is the pride and birthright of true French people :)

Several years ago I helped cofound a worker owned for-profit software and design cooperative. Its constitution was formulated around a mix of product incubation and consulting in support of labour solidarity through cultivation of progressively more interesting/motivating/challenging work. It was founded after we had all quit our high-performing and very profitable agency whose non-technical/non-creative ownership cut one too many corners on the sustainable systems that we were working on. Typical anti-flow type management decisions with arbitrary deadlines, death marches, and silly promises to clients despite solid contracts (which I had written and negotiated directly with the clients and our lawyers).

In the new co-op that us engineers and designers from the old company formed, there was growth early on, and a plan to expand to new cities/markets because the cost structure was horizontally scalable by design. This was before boutique remote app studios and incubators became commonplace (at least before all the really big agencies tried to get in on it, e.g. Sid Lee, Ogilvy).

Lo and behold, it turned out there were just barely enough members who wanted to take a more employee/entitlement oriented tack (as opposed to entrepreneurial) vs the other half. These members were of the same local cultural and linguistic background (the socially dominant one), whereas the rest of us were from all over. The locals formed a "cabal" of sorts and pushed for spending lots of money on a swanky office, vetoed horizontal expansion and denied the right to work for one of the members who had just moved to a new city as part of the previous plan (by voting to refuse to allocate some of our mandated work to him).

The original vision fell apart and was replaced by something more parochial. The more "ambitious" cofounders all left and the co-op continues to thrive in its own way today but as a shadow of the dream we had in the beginning.

An autocracy would not have endured these trials so early on. In top-down mode, the boss gets to enforce focus. Certainly not good if the dictator is not benevolent, but the experience provided for a lot of reflection on governance.

Reminds me of this explainer video I saw recently from CGP Grey: https://www.youtube.com/watch?v=rStL7niR7gs&t=344s

No matter the exact loci of power, there are always gradients between the points, and forces that can bring imbalance are very hard (perhaps undesirable) to control.

  "Some products are inherently unsafe, and that is precisely what the costumers want."

Swedish snus is made to minimize tobacco-specific nitrosamines (TSNAs), but seems to not address Polonium-210 on account of the radioactive stuff being "comparable to that from the natural background radiation sources or dental x-rays": https://harmreductionjournal.biomedcentral.com/articles/10.1...

That excuse doesn't seem to hold up to scrutiny. Who really wants to have 10-20 chest x-rays done per day? Are "dental" x-rays really as weak as "background" radiation?

Alcohol seems to be in a similar position - see the popularity of nice wine and beer over e.g. guzzling cheap spirits.

I think people want health AND pleasure, and are willing to do their own hedonic calculus.

Heard about it on a podcast recently, haven't had a chance to try.

- Request parsing

- Content negotiation

- Creating/rendering responses

- Subclassing or decorating custom views and/or viewsets

- Model mixins, CRUD glue

- Routing

- Content-type parsing and rendering

- Serialization/deserialization

- Validation

- Authentication

- Filtering

- Pagination

Personally, I found it was harder to get from zero to anything when having to slog through that level of verbosity. I'd rather save brain cycles for harder problems.

It was this experience on the job that led to the design of a better approach with Ramses: http://ramses.tech. You can also generate docs, tests and client libraries this way. If you like the look of that, I will make you an end-to-end prototype for free to help drive home the concept.

Some other awesome projects along the same lines are http://postgrest.com/ and https://flask-restless.readthedocs.org/en/stable/

It lets you generate end-to-end tests from a spec file for APIs that are themselves generated from the same file. Same for client libraries and API docs, generate all the things :)

I was thinking more along the lines of having contract breaches being publicly visible, which would motivate the more sociopathic egotists of the world to keep up appearances of fairness. Making it harder to just move to the next victim because a reputation could be built up.

Edit: also bridging the gap between expensive lawsuits that end up not being worth it, or the temptation to publicly shame the bullies which is almost always tacky and never dispassionate.

That's super interesting!

I wonder what kinds of stuff you could do with that in a fancy datastore like Elasticsearch or GraphDB.

If it could be mapped to a resource model in a straightforward way, it'd be fun to make a special "ORM" that plugs into an API framework to expose public services.

Which ones should I start with? Where to look in the docs to get started?

I would love to ask clients to "sign here" on a blockchain-enforced contract before they decide not to pay just because they feel like they can, despite perfect delivery to their own satisfaction.