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 Egypt's Giza pyramids: Radar finds underground structures
I would sincerely want to see more post and discussions on HN about this.

Spending 200K man years for a tomb seems like a pretty poor use of resources.

javierluraschi··on AI will take your job. Get over it
More like, AI will take your job and give you a new one.
javierluraschi··on AI will take your job. Get over it
Ugh, there is a lot of hype in the short term, probably not enough hype long-term, 5-10 years.
javierluraschi··on Using LLMs to Enrich Datasets
I like this approach, just a bit too expensive for large datasets and local LLMs not that great and harder to scale, this needs to mature a bit more but gets my thumbs up.
javierluraschi··on YC alum LunaJoy got $6.6M, offers job below minimum wage
Why? They are competing against other companies that are outsourcing the same job and probably paying less.
javierluraschi··on Ask HN: If you could start over, hardware or software?
Whichever one you like the most.

Hardware is 10x harder, more costlier more expensive mistakes, but also 10x less people tackle that space, so net net is all good, I’d pick whatever interests me the most

javierluraschi··on Vibe Coding a Product in 6 Days
That's perfectly fine as long as that algorithm has lower running time that writing code from scratch
javierluraschi··on Vibe Coding a Product in 6 Days
It only produces low quality if you don't code review the output and don't use subsequent prompts to refine.

In my case, I notice the extension was requesting more permissions that it should. This would have delayed approval on the Chrome store, so I directed the LLM to simplify permissions and the like.

We are still the ones accepting pull requests and asking for improvements, so vibe coding does not need imply low-quality at all.

javierluraschi··on [dead]
Is going to be really hard to make the government more efficient.

Might be better to create 2.0 agencies. Say, IRS 2.0, USPS 2.0, NASA 2.0, DOD 2.0, start them from scratch and then eventually shut down the 1.0 agencies.

Transforming existing agencies inside out, might be close to impossible.

The Innovator's Dilemma would likely be a case on why it's really hard to transform existing organization, it's possible, but extremely hard; better to let startups disrupt existing companies.

javierluraschi··on Ask HN: What Are You Working On? (February 2025)
Working on hal9.ai -- Long term, a Roblox for AI; short term, a Python customizable ChatGPT that is enterprise ready. Think of ChatGPT without the LLM and support for writing your own RAG.

https://github.com/hal9ai/hal9

javierluraschi··on Classic Data science pipelines built with LLMs
For those interested, you can use LLMs to process CSVs in Hal9 and also generate streamlit apps, in addition, the code is open source so if you want to help us improve our RAG or add new tools, you are more than welcomed.

- https://hal9.ai

- https://github.com/hal9ai/hal9

javierluraschi··on GitHub Status – Incident with Pull Requests and Issues
GitHub down
javierluraschi··on TikTok says it is restoring service for U.S. users
If your cloud provider tells you they are shutting you down on date X, you want to fight as hard as you can until X and then shutdown gracefully to have a chance to explain to your users why your system is going down. If you wait until you get shutdown, you have no way of pushing a graceful shutdown anymore.
javierluraschi··on TikTok says it is restoring service for U.S. users
https://www.reuters.com/technology/oracle-prepares-start-shu...
javierluraschi··on The thousand brains theory of intelligence
Re-reading this 2019 article, makes me think that Numenta's / Jeff theory alluded to the latest test-time compute paradigm in LLMs.
javierluraschi··on Show HN: Atlas of Space
I would love to be able to zoom out beyond solar system and grasp a sense of how far we are from other stars and our place in our galaxy, etc.
javierluraschi··on Salesforce will hire no more software engineers in 2025, says Marc Benioff
They are probably struggling to keep sales number up, they need more sales and cut costs. So makes sense to hire more sales people and freeze development… but that’s a sad story, let’s fix that sad story with AI:

AI is so good that everything can be automated, except for sales which needs a personal touch.

Wall Street happy, bonus for all execs, the end.

javierluraschi··on Show HN: A minimalist habit tracker with a wheel-based visualization
Good to see single page projects, can you try hosting it somewhere like GitHub pages?
javierluraschi··on Hal9: Create and Share Generative Apps
Connecting to a CRMs and building a generative application is supported, think "Chatting with your CRM" or even doing more complex tasks like "Analyzing your CRM by generating dynamic data dashboards".

To interoperate with other systems, every generative app built in Hal9 also comes with an API. So whenever the CRM has new data, it can call Hal9 to process records or do other kind of actions. We've seen some customers using tools like Zapier in combination with Hal9 to do more complex triggers, say, support SMS and the like.

So short answer, yes! But we expect some integration to happen with other systems in the form of APIs and triggers.

javierluraschi··on Hal9: Create and Share Generative Apps
Hi there HN! We built Hal9 to make it radically simpler to create, deploy, and share applications powered by LLMs, diffusers, and other AI models. Whether you're working on chatbots, agents, APIs, or generative apps, Hal9 is designed to minimize the engineering overhead so you can focus on the AI itself.

* Why Hal9?

Most generative AI projects end up dedicating the majority of their time to engineering challenges -- building interfaces, integrating tools, and managing infrastructure -- rather than focusing on the core AI work like refining prompts, implementing RAG strategies, or optimizing model performance.

Hal9 shifts that balance by drastically reducing engineering overhead. It offers a simple, lightweight interface built around Unix IO conventions like stdin and stdout, allowing you to focus entirely on AI innovation without the need to learn complex frameworks or deployment workflows.

With Hal9, you can prototype and run locally without extra dependencies, use our free online platform for quick deployments, or scale effortlessly to enterprise-grade solutions. We can also support organizations by enabling cloud deployments in their own environments or providing additional compute resources for enterprise customers.

Hal9 is designed to get out of your way so you can focus on building smarter, faster.

* What is Hal9?

Hal9 is a deployment platform purpose-built for generative AI, enabling you to create and deploy generative (LLMs and diffusers) applications (chatbots, agents, APIs, apps) in seconds. Key features:

Flexible: Use any library, and any model.

Intuitive: No need to learn app frameworks, simply use `input()` and `print()`.

Scalable: Designed to integrate your app with scalable technologies like Docker and Kubernetes.

Powerful: Using an OS process (stdin, stdout, files) as our app contract, enables long-running agents, multiple programming languages, complex system dependencies, and running arbitrary code in secure Kubernetes pods.

Open: The code behind the Hal9 app, is also open source and open for contributions under our repo.

* The Philosophy

We believe the Python ecosystem already provides great libraries for everything from LLM interactions to generative tasks. Instead of reinventing those wheels, Hal9 integrates them into a unified workflow, letting you focus on AI-specific challenges like retrieval-augmented generation (RAG), fine-tuning, alignment, and training.

Hal9 is perfect for developers who want to experiment, iterate, and deploy AI apps quickly without becoming mired in engineering tasks like frontend design or backend integration. It's also ideal for teams looking to collaborate, thanks to the open architecture and straightforward app structure.

* Our Journey

We started Hal9 in 2021 with the goal of simplifying AI development. Initially, we focused on web developers, combining AI with technologies like D3.js and TensorFlow.js. While the low-code interface was popular, users wanted that, but with Python support.

In 2022, we took less-code a step further and embraced LLMs like GPT-3, moving towards automatic code generation and simplifying the UX. After several iterations, Hal9 has evolved into a platform that enables faster, easier AI app development.

* Resources

We are actively publishing posts that demonstrate how to integrate your favorite frameworks with Hal9. Here are some of the technical blog posts already available:

Hal9 with OpenAI Swarm: https://hal9.com/docs/blog/swarm-openai

Hal9 with NVIDIA NIM: https://hal9.com/docs/blog/nvidia-nim

Hal9 with Dagworks: https://hal9.com/docs/blog/burr

Hal9 for Text-to-SQL: https://hal9.com/docs/blog/txt-to-sql

Let us know your thoughts, feedback, and ideas -- Hal9 is as much about building apps as it is about creating a community of creators.

javierluraschi··on Independent directors of 23andMe resign from board
If we agree that public market are the closest we can get to realtime market value, it's hard to argue it's easy to find a better offer. Otherwise, many other companies that are down 95% from their 2021 highs should just find a buyer of their companies to maximize shareholders value.

I think we are disappointed that the market is giving $ME and many other companies non-ZIRP valuations at 95% discounts, and we don't have the cash to buy them at the bottom, like she can.

javierluraschi··on Independent directors of 23andMe resign from board
I'm just arguing that she can't let the company go bankrupt when she placed a bid for $0.40. Even if she could, she could loose control of the company when all shareholders get wiped out, not sure the court will let her keep 70% through bankruptcy.
javierluraschi··on Independent directors of 23andMe resign from board
She (the CEO) has a fiduciary duty to act in the best interests of the company and its shareholders. Intentionally driving a company into bankruptcy for personal gain is a massive breach of that duty and opens her up to all kinds of legal trouble. So I would think that's a very bad idea.
javierluraschi··on FGA4 cause Google Analytics 4 sucks
We used to love Google (Universal) Analytics, then came GA4 and it's no bueno.

We are launching https://fga4.com to bring back the reports we lost.

The new reports are built with open source technologies like Python, Pandas and Streamlit to make sure that this never happens again.

Connect with us at https://twitter.com/F__GA4

javierluraschi··on Weaveworks is shutting down
What's your ideal B2B sales strategy that would work for you?
javierluraschi··on Essential AI
Pre-vision, pre-team, pre-product, and pre-revenue -- An opportunity to build the future in this blank canvas with the founding team.
javierluraschi··on Essential AI
Looks like https://hal9.com as in enterprise data analytics with LLMs.
javierluraschi··on Ask HN: Why is ai.com redirecting to x.ai
I find it quite faster to type than chat.openai.com
javierluraschi··on Ask HN: Why is ai.com redirecting to x.ai
Not sure if he has that much leverage... maybe he agreed to sell to OpenAI access to Twitter/X data only if the domain was part of that deal?
javierluraschi··on Ask HN: Why is ai.com redirecting to x.ai
Here is a reference but not much information about what happened nor why: https://analyticsindiamag.com/musk-buys-ai-com-from-openai/
← PreviousPage 2 of 7Next →