HNHacker News
TopNewBestAskShowJobs

javierluraschi

402 karma · joined March 23, 2014

Founder at hal9.com working on cloudish.ai

Previously RStudio, The Linux Foundation, Microsoft Research, and SAP.

x.com/javierluraschi github.com/javierluraschi linkedin.com/in/javierluraschi

submissionscomments
javierluraschi··on Show HN: Hal9 – Low-Code Analytics for the Web
Hello HN from the Hal9 team (https://hal9.com). We’re building a low-code analytics platform based on web-technologies focused on making AI more accessible.

We have worked many years with different ML technologies across many projects, feeling frustrated at the amount of engineering effort required to set up even simple AI projects.

So with Hal9, we are lowering the barrier to AI by providing an integrated platform that can help you build data transformations, visualizations, and predictions from a library of reusable components. You drag-and-drop these components to compose data pipelines, but you are never stuck with us since you can always export the pipelines as JavaScript, R, Python or create your own blocks.

How is it different from existing solutions? Most AI applications require integrating a web frontend with an AI backend usually running Python, PyTorch, TensorFlow, or h2o. One can build the AI backend in Docker, Kubernetes, MLflow, and such; you then set your web stack with say, NodeJS and React or go for a JAMStack. Between them, lies a REST API. This is fine for large projects, but setting this infrastructure, defining the API and maintaining this system is usually quite complex and expensive.

Hal9 solves this problem by integrating the frontend and backend under the same platform, giving you AI templates to start from, and letting us worry about keeping your application running. In many cases, all you have to do is drag-and-drop, in others you write code blocks in JavaScript, NodeJS, Python or R and we deploy the frontend and backend for you.

We found out web developers, business experts, and data scientists can benefit from Hal9. In teams that have all these disciplines, Hal9 becomes the place where they can collaborate from prototype to deployment. For teams that are missing one discipline, Hal9 fills that void by providing ready-to-use templates, blocks and a worry-free infrastructure.

In the short term, while you can do some basic AI like running image classification, regressions and sentiment analysis; we’ve put most of our effort in data transformations and visualizations in the browser. This helps us reduce cloud-compute costs, increase privacy and speed, at the cost of working with datasets of at most 1M rows. It was important to us to have strong building blocks and be able to scale to millions of users, even if it meant having to wait a little longer on proper AI using large datasets. That said, we already have early support for Python and R, and we will keep growing the platform and team to support larger datasets.

In addition, we have an open-core technology and our blocks are open source; so we are also hoping that experts in the community will create components in our GitHub repo for others to reuse. Check out https://github.com/hal9ai/hal9ai to contribute, submit issues, questions, or suggestions.

We’ve come a long way since our alpha release half a year ago (https://news.ycombinator.com/item?id=27334145), and we are very excited to see what you can come up with on the platform. Let us know what you think!

javierluraschi··on Musk Asks Twitter: Should I Sell 10% of My Tesla Stock?
Well played. No: Permission to be against new capital gains tax. Yes: Permission to sell at ATH.
javierluraschi··on My first experiment in generative art – Built with JavaScript and SVGs
Nice! We’ve been playing with some generative art as well here at Hal9, would appreciate feedback! Hal9 is a data visualization platform but we’ve find it useful for generative art as well, generative art is quite addicting.

https://news.hal9.ai/

javierluraschi··on Is BI dead? – On dismantling data's ship of Theseus
Would love to hear your feedback for https://hal9.ai -- We are building an open source platform for data analysis based on reusable code blocks and a community to build and monetize their contributions. We are pretty early in our journey, launched our alpha and getting ready for our beta release, but would love to hear your thoughts. You can find me at javier at hal9.ai. Cheers!
javierluraschi··on Data Science with JavaScript: What we've learned so far?
Author here, we can save some clicks by reposting content as comment:

Hi there, since the beginning of this year we've been exploring how far we can take Data Science with JavaScript. As part of this journey, we started hal9.ai [1], an integrated environment to help us be more productive when analyzing data with JavaScript.

We want to ask for your feedback, but more importantly, we want to use this post to share what we've learned so far:

1. Visualizations: JavaScript is great at visualizing interactive data, this is probably obvious but worth mentioning nonetheless. Some of the highlights here, D3.js [2] is still a great library to perform visualizations; however, D3.js is really low level -- Kinda like TensorFlow [3], not Keras [4]. We actually went to create our own charting library to combine the flexibility of D3 with the ease-of-use of other libraries like Plotly [5]; just to find out later on that Plot.js [6] got launched as an amazing library that builds on top of D3. So we ended up integrating Plot.js as our recommended charting library.

2. Transformations: We found out that JavaScript in combination with D3.js has a pretty decent set of data import and transformation functions; however, it comes nowhere near to Pandas [7] or dplyr [8]. After shopping around, we found out about Tidy.js [9], loved it, and adopted it. The combination of Tidy.js and D3.js and Plot.js is absolutely amazing for visualizations and data wrangling with small datasets, say 10-100K rows. We were very happy with this for a while; however, once you moved away from visualizations into data analysis, we found out 100K rows is quite restrictive, which is also slow when having 1K-10K columns. So we switched gears and started using Arquero.js [10], a columnar JS library that enabled us to process +1M rows in the browser, decent size for real-world data analysis.

3. Modeling: We are currently exploring this space so our findings are not final, but let us share what we've found so far. TensorFlow.js [11] is absolutely amazing, it provides a native port from TensorFlow to JavaScript with support for CPU, WebGL [12], WebAssembly [13] and NodeJS [14] backends. We were able to write an LSTM [15] to do time series prediction, so far so good. However, we started hitting issues when we started to do simpler models, like a linear regression. We tried Regression.js [16] but we found it falls short, so we wrote our own script to handle single-variable regressions using TF.js and gradient descent. It actually worked quite well but exposed a flaw in this approach; basically, there is a lot of work to be done to bring many models into the web. We also found out Arquero.js does not play nicely with TF.js since well, Arquero.js does not support tensors; so we went on to explore Danfo.js [17], which integrates great with TF.js but we found out it's hard to scale transformations to +1M rows and found other rough edges. Since then, well, we started exploring Pyodide [18] and perhaps contributing to Danfo.js, or perhaps involving more server-side compute with NodeJS, but that will be a story for another time.

So net-net, we are still super excited about exploring Data Science, Data Engineering, Visualization and Artificial Intelligence with JavaScript; but realistically, it is going to take a few years for this to mature.

In the meantime, we think Data Science with JavaScript shines with smaller datasets and interactive visualizations; which we believe Hal9 can help you be productive at. That said, we do believe that motivated JavaScript users can help unblock themselves by adding new functionality and contributing back libraries to NPM or components to our open source project, please do reach out in Hal9's GitHub repo [19] if you wanna lend a hand!

Alright, so call to action? Please head to hal9.ai and give it a shot! We would love to hear where you think this could be useful, what features we are missing, and any feedback you may have.

To keep in touch, please subscribe to our weekly email at news.hal9.ai [20], contact us at info@hal9.ai, or follow us on Twitter as @hal9.ai

Thanks for reading along!

[1] https://hal9.ai [2] http://d3.org/ [3] https://www.tensorflow.org/ [4] https://keras.io/ [5] https://plotly.com/ [6] https://observablehq.com/@observablehq/plot [7] https://pandas.pydata.org/ [8] https://dplyr.tidyverse.org/ [9] https://pbeshai.github.io/tidy/ [10] https://uwdata.github.io/arquero/ [11] https://www.tensorflow.org/js [12] https://get.webgl.org/ [13] https://webassembly.org/ [14] https://nodejs.org/en/ [15] https://en.wikipedia.org/wiki/Long_short-term_memory [16] https://github.com/Tom-Alexander/regression-js [17] https://danfo.jsdata.org/ [18] https://pyodide.org/en/stable/ [19] https://github.com/hal9ai/hal9ai [20] https://news.hal9.ai/

javierluraschi··on China has forbidden under-18s from playing games for more than three hours/week
Where is the line compared to government enforcing parents to not leaving their kids alone in a car or home, or sending their kids to school?

How do we measure what level of control we have/deserve on our kids?

I'm not disagreeing, just arguing that there are cases where one could argue that the government is overstepping on how we parent our kids.

javierluraschi··on Software Engineer launches startup despite long Covid
GeekWire wrote a post on my personal story with long COVID while launching our startup. There is definitely a lack of awareness, resources and research on long COVID. I must say I felt like I was going crazy for a while there until I got a proper diagnosis. I'm hoping the word gets out and people start taking this seriously. I was lucky enough to not get the wost version of long COVID, but others out there are suffering debilitating symptoms that have taken their well-being overnight. If you want to chat on a personal level, find me at javier at hal9.ai
javierluraschi··on Grand jury indicts Trevor Milton, Nikola founder, on three counts of fraud
Although, most V1 software is released early, as an MVP (Most Viable Product), get a lot of feedback and iterate. So it's a very reasonable approach to release FSD incrementally and get early adoption and early feedback. Again, it's a pretty hard problem to solve; so even if they fail, lots of respect from me to their team for getting this far.
javierluraschi··on Grand jury indicts Trevor Milton, Nikola founder, on three counts of fraud
The possibility is not enough. They have to make FSD work in older Teslas to fulfill the original promise, somehow.

That said, if it doesn't happened this year, it doesn't mean it will never happen.

The first problem they need to solve is FSD. If they can't solve FSD even with latest hardware, then well, that would be bad on its own.

The question is, what happens once they solve FSD, do they port? Maybe, I don't think it will be nearly as hard. Now realistically, it's probably cheaper to give free upgrades to everyone and not have to do the port. To me that feels fair, is like "hey, I promised you FSD on HW1, is gonna be a pain to port, but here, take this free HW upgrade", I'd be OK with that.

Now if they don't compensate HW1 users to some degree once FSD is solved, I do feel that would not be cool.

But my point here is that, we are jumping to conclusions, we don't know yet. We can't claim (yet) that FSD will or will not happen.

javierluraschi··on Grand jury indicts Trevor Milton, Nikola founder, on three counts of fraud
That's a bit harsh. Many technology products have delays, it sucks, makes customers unhappy, but is not uncommon and is not considered lying, not in my book at least. Windows Vista was delayed for many many years [1]. They actually promised a full database replacement to their file system that it never really shipped, etc. This is not to criticize them, but just to say that when ambitious goals are set, it's often the case that delays happen. You can definitely make the case that maybe it's better to release products Apple-style and just unveil them when they are ready, that would be a fair point.

[1] https://en.wikipedia.org/wiki/Development_of_Windows_Vista

javierluraschi··on Grand jury indicts Trevor Milton, Nikola founder, on three counts of fraud
I'd add, we can be more specific about what a "case" is. I'd say, a trip might describe this better. As in, 90-99% of trips FSD Beta 9 will take you there without interventions. That's still lame, but still quite impressive. Achieving 90-99% accuracy on trips still seemed impossible one year ago, I don't think that's the case anymore. The remaining risk is, how fast can this improve between 99% to 99.9% and then to 99.99%, etc.
javierluraschi··on Grand jury indicts Trevor Milton, Nikola founder, on three counts of fraud
Correct! I'm saying FSD only works on 90%-99% of cases, which is currently insufficient to be actually considered FSD. Keyword, currently.

However, I'm NOT stating that this won't ever work. I do believe that FSD will work in the future. Are there risks? For sure. But I'm very optimistic.

I'm trying to communicate that, even with current in-car hardware, this should be sufficient. Why? Cause driving the car with a neural-network is actually quite trivial. Now, training the neural-model is actually REALLY hard. But the training does not happen in the car, car just collects data and sends it to a data center with GPUs that trains the network. If any hardware will need upgrade, is the data center were training happens, not the actual in-car hardware.

Are you really saying that FSD will never work? Cause that's a pretty bold claim.

javierluraschi··on Grand jury indicts Trevor Milton, Nikola founder, on three counts of fraud
Replacing hardware does not make old hardware incapable of FSD.

To give you an example, ImageNet for image classification required for many years a descent computer and GPU to run inference. Today, an improved version called MobileNet, runs in phones with no GPU from advancements in AI training with half-precision and why not.

So, how can we be so sure that FSD can't EVER be ported back to HW1?

javierluraschi··on Grand jury indicts Trevor Milton, Nikola founder, on three counts of fraud
AI algorithms are split into training and inference.

Training is the process of generating the parameters for the neural network, it usually takes significant resources to train and is a very hard process.

Inference is the process of using a pre-trained neural network to perform predictions.

Tesla's car have hardware to perform inference. From working on the field for several years, I do believe that they have hardware capable of performing inference (driving a car).

However, the hardware that trains the neural network is comprised of hundreds, if not thousands or tens of thousands of GPUs, that hardware lives in Tesla's data centers (or some other cloud provider).

The really hard part of this project is on training, not inference. So while there is some risk that the car (inference) hardware might require an upgrade, this seems very unlikely to me.

javierluraschi··on Grand jury indicts Trevor Milton, Nikola founder, on three counts of fraud
Historically, there have been significant bottlenecks on algorithms, I'd argue software plays a bigger role than hardware; especially during inference.

Going back to AlexNet, while GPUs helped the training process, it included significant software improvements that made a huge difference: training with dropout, convolutional neural networks, using rectified linear units, etc.

Also, training hardware is much more relevant than hardware for inference. Tesla's cars do not include hardware to train the neural network but just for inference. The training hardware is actually a server cluster that Tesla owns. So if hardware is insufficient, Tesla might have to buy a larger training cluster, but not necessarily upgrade the inference hardware.

That say, I do agree that there is some risk in requiring a hardware upgrade for inference, but to me that seems very unlikely.

javierluraschi··on Grand jury indicts Trevor Milton, Nikola founder, on three counts of fraud
What's wrong with the latest statement?

Technically all cars have a proper GPU (Tesla's "GPU") and 360 cameras, the reliability is probably in the 90%-99% (which, yes, is pretty terrible to be considered fully autonomous, but impressive from a technical standpoint). Technically is correct to say that all they have to do is to "improve the software". That will be a lot of work, sweat and tears, but still correct.

javierluraschi··on I want to learn D3. I don’t want to learn Observable. Is that ok?
I sympathize with the comments, as a software engineer, this comment from Mike really describes my frustration with Observable's runtime:

> Observable notebooks are like spreadsheets, where cells run automatically whenever you edit or values change. That’s not how conventional (imperative) programming languages work, so sadly you can’t simply copy-paste a whole notebook into a vanilla JavaScript application.

However, I do think many others will prefer their reactive runtime; so this is by no means a criticism to Observable, they are doing great work. They are just not targeting JS-purists and that might be the right call.

As a software engineer, I love D3 but don't want to be stuck with a reactive runtime that is not vanilla JavaScript.

I got myself to build https://hal9.ai, an integrated environment to do data analysis based of JavaScript. Is not a D3 learning environment but you can certainly use it for that purpose. If someone is interested in providing feedback or helping with D3 examples, please do so at https://github.com/hal9ai/hal9ai or shoot me an email at javier at hal9.ai

javierluraschi··on D3 7.0
Why can't we have both? -- I had a similar frustration and got to work on https://hal9.ai, you get prebuilt blocks to easily import/transform/visualize data, while also being able modify the D3/Plot sources to your hearts contempt.
javierluraschi··on JavaScript and the next decade of data programming (2020)
Couple days ago we posted a Show HN with similar arguments on why JS for data makes sense, see: https://news.ycombinator.com/item?id=27334145 -- Ben's article is much more eloquent and presents several examples, but there are a few libraries worth complementing this article: danfo.js (dataframes with tensorflow), ml5.js (a keras-like library), tidy.js (a tidy port to JS), and tensorflos.js.
javierluraschi··on Show HN: Hal9 – Roblox for AI
Thank you! Will work on those in the next day or so.

If you feel like, would love to keep in touch by email (javier at hal9.ai), at the very least, I can give you free access to the proper alpha release.

javierluraschi··on Show HN: Hal9 – Roblox for AI
Thanks! No, we have not but we will take a look at Google Cloud Run, I agree is quite painful to wait for this.

Great suggestion. Yeah, we don't use "Roblox for AI" in our website, but was hoping the HN community would get the analogy.

javierluraschi··on Show HN: Hal9 – Roblox for AI
Thanks for playing with this, really appreciate it!

Yes, so currently our budget is low and we turn on AWS EC2 workers on demand... so most likely, "Pipeline running" triggers as we wait for about 90s for the EC2 instance to start. We should be able to keep machines on longer as we get our first few customers.

Thanks, we do support fetching data from JSON, is there a difference with "web JSON"? XML/RSS would be nice indeed, created this issue: https://github.com/hal9ai/hal9ai/issues/4

javierluraschi··on Show HN: Hal9 – Roblox for AI
Good point. I don't see many AI apps being experienced together in the same way Roblox games are experienced together. The social aspect for Hal9 would be on developing AI solutions together, building components together, etc; but I agree most users play Roblox socially while only a few build games together.

We considered other analogies like "Unity for AI", but we felt it missed the marketplace component. Another strong contender was "Unity and Steam for AI". At the end, these analogies are going to be imperfect but hopefully they give a quick (yet incomplete) sense of what is being built. In general, we don't use this analogy in our site, but we figured HN would appreciate the analogy.

javierluraschi··on Show HN: Hal9 – Roblox for AI
We believe everyone should be able to use and profit from AI, so we spent this year building Hal9, which aspires to become "Roblox for AI".

* What is Hal9?

Roblox is many things: A game development environment with a visual editor that makes it easy to get started but unconstrained to advanced users that want to code. Roblox is also a library of assets, and a marketplace for games.

We are taking a similar approach, we are building a visual editor to assemble anything from simple data analysis by importing data and visualizing it with charts, to data forecasting with regressions, to complex workflows to do web-scraping and run deep learning models like LSTMs.

However, visual editors are constrained and at some point they just don't work, so every block is backed by open source code which you can modify to your hearts' contempt using our code editor.

In addition, you can also add new blocks and share them with others in a library of assets. We are currently using GitHub as a library of open source scripts.

Finally, we have aspirations to help contributors profit from their work by sharing profits with them.

* How are we building this?

We are building everything with web technologies, let me explain why.

First, the whole purpose of this is to make AI more accessible. Technologies like Electron allow us to easily deploy to multiple platforms and avoid having users troubleshoot system dependencies. To put it simply, web technologies allow users to get started with no additional setup.

However, we also see the ecosystem of web technologies converging with Data Science and AI. For instance, the tensorflow.js [1] project allows users to run TensorFlow in the client and optionally train models in GPUs in NodeJS. We currently provide NodeJS infrastructure to run server-side code and hopefully GPUs as well soon.

Another great example is webassembly, which can allow us to run Python with PyIodide [2]; same for SQL with sql.js [3], etc. We are also very excited for WebGPU [4], an initiative to bring proper GPU support to web technologies, which should make it so much easier to use proper GPUs locally.

Momentum is growing around Data Science with libraries like tidy.js [5], danfo.js [6], and ml5.js [7]. Currently this ecosystem is a bit fragmented but we believe they will eventually come together.

Finally, web technologies like CSS and D3.js are arguably the best platform for visualizations. We are big fans of the new ObservableHQ Plot [8] library which we are already using in Hal9.

* Call to action

Head to https://hal9.ai and try it out.

You will need your desktop to interact with the embedded app at the top of our site. The app is diabled on mobile but the site should give you a good idea of how it works.

We are looking for feedback, users and contributors.

If you find this useful, you can download Hal9 for Windows, Mac, or use the Web version and our cloud service for $19.99/month.We are planning to use this revenue to become self-sustained and share profits with our community of developers.

As a developer, you can download and use the development versions to contribute to the open source repo of scripts in https://github.com/hal9ai/hal9ai

* Resources

- Site : https://hal9.ai - Github : https://github.com/hal9ai - Twitter : https://twitter.com/hal9ai - Discord : https://discord.gg/dKYgga84 - YouTube: https://www.youtube.com/user/jluraschi

* References

[1] https://www.tensorflow.org/js [2] https://github.com/pyodide/pyodide [3] https://github.com/sql-js/sql.js/ [4] https://www.w3.org/TR/2021/WD-webgpu-20210518/ [5] https://pbeshai.github.io/tidy/ [6] https://danfo.jsdata.org/ [7] https://ml5js.org/ [8] https://observablehq.com/@observablehq/plot

javierluraschi··on Show HN: Run Node.js remotely in the browser
Related to WebContainers, see https://news.ycombinator.com/item?id=27223012; however, since they are not yet available, figured out some might want to start prototyping solutions with at least Node.js remote execution in the browser.

Planning to add support for arbitrary libraries over the weekend.

Currently running in a shared docker container, so please don't use this for sensitive data.

javierluraschi··on Dataflow, a self-hosted Observable notebook editor
For viz/DS/ML/AI with JS/TS is either observablehq or and IDE with custom extensions; this project looks relevant if you are already into observablehq.

Shameless plug, we are building a few tools for JS to narrow down this gap as well: - https://hal9.ai (Drag&Drop / IDE) - https://marketplace.visualstudio.com/items?itemName=Hal9.hal... (VSCode extension) - https://observablehq.com/@javierluraschi/running-nodejs-in-o... (ObservableHQ extension)

Would love to chat if you are interested in providing feedback, I'm in javier at hal9.ai. Cheers.

javierluraschi··on JavaScript for Data Science
I'm glad more people are doing DS/ML/AI with JavaScript, thanks for this book and keep up the great work! -- We are also working in this space, would love to connect, you can find me in javier at hal9.ai
javierluraschi··on JavaScript for Data Science
A few reasons, https://venturebeat.com/2021/04/23/4-reasons-to-learn-machin...

Personally, I'm excited to build apps that don't require cloud computing and if they do, have access to one of the largest software engineering libraries through NPM. Sure, I'm not doing just Data Science in JavaScript but rather building apps that use DS/ML/AI, but that's still a valid use case. The alternative would be to use Python for prototyping then rewrite for production apps.

javierluraschi··on JavaScript for Data Science
Glad you also see it this way! Would love to chat with you and get some feedback on a platform we are building at hal9.ai, my email is javier at hal9.ai -- Looking forward to chat.
javierluraschi··on JavaScript for Data Science
My hunch is that there has been 10X more investment in engineering for JavaScript: nodejs, webassembly, webgl, webgpu, react native, deno, typescript, electron, chrome, etc. That will be harder to rewrite in Python than to rewrite TensorFlow and a few math libraries in JavaScript.
← PreviousPage 4 of 7Next →