Congrats on putting this service together.

Was reading some of the free articles on my phone (iOS/Safari) and I noticed that I’m unable to select, this lookup, words.

Is there any reason why you’ve disabled text selection on the main body of the articles?

If it is to stop people from copying it won’t help.

Anyway, it seriously degrades the UX that one cannot use the native phone features such as lookup. Especially when reading articles of this type.

I use alpakr a couple times a week for learning about new packages, crates, etc.

Pipe from stdin and output package summaries.

I wanted to simplify this process by making a common interface to look up packages from many common package repositories without having to care too much about format of the input and remembering CLI parameters.

The basic idea is that you copy paste a snippet of text from the console, from GitHub in the browser, or by piping to stdin into “Alpakr” and out comes a summary plus link to the package in the package repository.

    $ echo ‘
      "flow-bin": "^0.123.0",
      "husky": "^3.1.0”,’ | alpakr

    flow-bin -> Binary wrapper for Flow - A static type checker for JavaScript
    https://npmjs.com/package/flow-bin

    husky -> Modern native Git hooks made easy
    https://npmjs.com/package/husky

Once the packager is determined it is (mostly) trivial to extract packages names to do the actual look up in the packagers repository, eg. If ‘cargo’ then packages are looked up in crates.io, ‘composer’ is packagist.org, ‘npm’ in npmjs.com and so on.

All package details are transformed to a common format and stored/cached in a DynamoDB for future lookups.

It has been a fun project to work on covering subjects I wanted to learn and improve on: ML with scikit-learn, building images to run in AWS Lambda, running Rust in AWS Lambda (built on Mac M1), warming Lambdas, multi-region serverless, and more.

For me the working out the ML preprocessing function was a fun task. If you think about it is more the flow of symbols rather than the actual names/words that matters when you are only looking at snippet of the full document. Also not all documents are structured e.g I wanted it to support the stream of “Compiling regex v1.0.0” when building a Rust binary. Meaning alpakr supports copying text of that format as well.

What was not so fun was retrieving the training data. I would love to hear of some better approach/source. I searched for package files on Github and used the API to fetch the files. Took a long time.

I also enjoyed trying to make a small efficient model. I ended up with 150kb model that had a 98.5% accuracy against test data. Ironically the framework dependencies needed to run it is short of 300mb. But I’m guessing a small model also loads faster, so decided to keep it small.

In addition to installing the tool with npm you can install it with `cargo install alpakr-cli`.

See either npm or crate for examples of using the CLI.

I was thinking about adding a `—hint` function to override the ML part, but I kind of like how simple it is without. What do you think?

ps: I’ve also made a mobile app so you can read code in the phone, flip to the app that copies code from clipboard (opt-out possible) and looks up automatically.

Comments URL: https://news.ycombinator.com/item?id=30910584

Points: 3

# Comments: 1

Small thing that irks me. The app icon looks too much like a unicorn. A stocky little Pony would be so much nicer and true to the name.

This is not a package/parcel service. Carries will be packing the items themselves in their own luggage. Carrier and requester are responsible for reporting goods in customs. The service will support this with communicate managed information about each location and the service will we ranking and testimonial based similar to AirBnB and Couchsurfing.

We will eventually be providing an API to allow for blogs, apps and curator services to make use of it as well.

For people thinking: commercialisation of drug mules, silk road, etc. In stead of thinking that everyone are criminals, try think that they are not (because their aren't) and think about the benefits.

Comments URL: https://news.ycombinator.com/item?id=7384230

Points: 3

# Comments: 2