Rui Machado

Leading AI innovation. 2x O’Reilly Author on Analytics Engineering and Advanced SQL. Executive leader on Data and AI with 15 years of practitioner experience.

[Rui Machado @axemaster26]

@MaioLabs Não há melhor validação que a dada pelos profissionais para quem estamos a desenhar soluções

[MaioLabs @MaioLabs]

Enquanto a indústria debate o hospital do futuro, nós estamos a construí-lo.

Carlos Santos Moreira, Professor de Medicina na Faculdade de Medicina da Universidade de Lisboa, escreveu no SOL: "A grande oportunidade da IA não é automatizar o hospital atual. É permitir-nos desenhar outro hospital."

Na @MaioLabs é exatamente isso que fazemos. Neste vídeo, um primeiro olhar sobre nossos produtos.

A história completa vem aí. 🏥

[Rui Machado @axemaster26]

https://t.co/E9cE2XexWn

[Rui Machado @axemaster26]

60 GB of Parquet on a 16 GB laptop means renting a bigger box or standing up Spark to aggregate it.

q.collect(engine="streaming")

Polars runs it in batches, so nothing has to fit in RAM. Ops that cannot stream fall back to in-memory.

Learn more: https://t.co/2XwSjVaBHc

[Rui Machado @axemaster26]

Check out my latest article: The Blanket Effect: why your brain ignores what you paid for and falls in love with what came for free https://t.co/7vBt2Nt1zg via @LinkedIn

[Rui Machado @axemaster26]

https://t.co/ek8gtgMcxU

[Rui Machado @axemaster26]

Did you know some inference frameworks stream model weights instead of loading everything into GPU memory?

This lets you experiment with larger models on existing hardware. The trade-off is slower inference due to disk I/O.

AirLLM is one example:
https://t.co/9xWnIVfksj

[Rui Machado @axemaster26]

If you are worried about your coding agents LLM inference costs, there are many options from you.

Take a look at:

- RTK: Cache the most common dev commands to that the llm doesnt beed to be triggered: https://t.co/ulKheB4JzI
- LLM Proxy: Connect to many LLM providers and design a routing mechanism that based on an intent-detector routes your prompt to the most fitted and optimized llm https://t.co/CKMwSleuly
- Pragmatism: Do you really need an LLM for your use case? Do you need a frontier model or an small open source one is enough?

[Rui Machado @axemaster26]

@sattyyouneed Try rtk for caching your frequently prompt responses https://t.co/dL469jrtyw

[Rui Machado @axemaster26]

@Prathkum But still decisions are being made by humans :) So the real question is who is creating this crazy world ?

[Rui Machado @axemaster26]

@pvergadia We are all distilling each other now :)

[Rui Machado @axemaster26]

@Im_IrushiK It's the old closed versus open weights perspective. When you subscribe
OpenAI or Anthropic you get the LLM wrapped on a huge harnessing layer. But the gap is closing

[Rui Machado @axemaster26]

@kylejeong Funny enough, better context in an average model often beats low context on a better model. You can get better answers while saving on tokens

[Rui Machado @axemaster26]

Did you know structured outputs are only half the solution?

Always validate the response against a schema before using it.

A valid JSON isn’t always valid business data.

https://t.co/JRyO17YvtQ

[Rui Machado @axemaster26]

@Alibaba_Qwen OpenWeights models are gradually winning their place! Everyone that cares should start experimenting with it

[Qwen @Alibaba_Qwen]

Meet Qwen3.8-Max: A New Bar for Coding and Cowork.

[Rui Machado @axemaster26]

Did you know Ollama + a downloaded model from HuggingFace + enough RAM can run an LLM locally with no cloud API dependency after setup?

Better privacy, lower recurring costs, and full control over your inference pipeline.

https://t.co/KkQviH4vdj

[Rui Machado @axemaster26]

New book is out.

This time we covered advanced topics such as how can you leverage SQL-based technologies to build data science applications, train GenAI models and even start your knowledge graph journey.

https://t.co/w1jeMwu0hu

@OReillyMedia

[Rui Machado @axemaster26]

@amasad Amazing @amasad. Wonder how you worked the adoption and evaluation of this with decision makers. Not eval of the technical flow, but rather the consumers of it, did you measure the benefit already?

[Amjad Masad @amasad]

We built a self-driving & self-correcting shared semantic layer on top of our databases, conversations, and docs.

Everything is queryable & joinable—regardless of source!

So now anyone at Replit can ask questions that previously needed a team of data scientists weeks of work.