Pedro Vilas-Boas

This week, we started a new course at Universidade Federal de Itajubá, IESTI05 - Edge Machine Learning Systems Engineering. IESTI05 is a 15-week undergraduate course designed to teach students how to implement AI systems on edge devices, specifically using Raspberry Pi platforms. The course is based on the e-book: "Edge AI Engineering" by Prof. Marcelo Rovai, UNIFEI 2025 (https://lnkd.in/eRrvn9FF). Open materials will be uploaded to this repo (https://lnkd.in/eRp2Q63A) every week. - Slides, Notebooks, Code, and Docs in English - Videos in Portuguese #EdgeAI, #AI

124

3 comentários

I've found that the Superpowers framework (https://lnkd.in/dMd5sHz9) is very effective when working with coding agents. I think most engineers working with coding agents realise how important the planning phase is. However, I've never been *completely* convinced by the native planning mode that you get with tools like Claude Code or Cursor. Over the last couple of weeks, however, I've gotten pretty consistent planning documents from Claude via the Superpowers plugin. I also like how each plan is broken down into two separate documents: one for design, and one for implementation. The actual implementation is then carried out via a new subagent per task, with additonal spec compliance and code quality reviews in between. This does obviously result in higher token usage, but I've found that the results are generally consistent (and I don't have to constantly keep going back and forth). Definitely worth taking a look at!

27

Our COO Margarida Garcia addressed a critical challenge at MIT Technology Review's EmTech conference: why context is everything in AI coding. The problem? Current tools stay very much at the surface. They can fix isolated bugs but struggle with multi-file editing and understanding how complex systems connect. When errors span interconnected files, generic AI tools can't trace the root cause contextually. "Understanding a codebase in some enterprise environments is effectively a nightmare," Garcia explained. "Having tools that bridge that gap and actually ingest, learn from the context—that can be your intelligence layer that knows more about your codebase than any engineer around." This is why poolside deploys within your security boundaries, understanding your specific context rather than offering surface-level solutions.

262

8 comentários

🎁Chasing Excellence: FEP Open Challenge🎁 In the FEP tools landscape, mature commercial software—with its complete toolkits, exceptional stability, and proven accuracy—has long set the benchmark. Yet, their steep costs not only complicate the cost–benefit analysis compared to experimental testing but also slow down the integration of AI-driven innovations in developing next-generation tools.😳 This reality inspired us to develop Uni-FEP.💪 Since Uni-FEP's first launch in 2021, over 200,000 FEP calculations have been performed by commercial users on the Hermite®️ platform, affirming our confidence in Uni-FEP’s stability and accuracy. 👍 However, as a new player in the FEP market, Uni-FEP’s true potential remains largely unproven to many.🥺 To showcase what Uni-FEP can really do, we’ve released the Uni-FEP-Benchmark—a collection of over 100 test cases covering a wide range of protein targets and diverse chemical modifications. This benchmark not only demonstrates Uni-FEP’s performance but also signals our unwavering commitment to developing a tool that truly works.💡 Understanding that many in the community may still have concerns about Uni-FEP’s capabilities, we’re launching the FEP Open Challenge (https://lnkd.in/g3ZffqXM). We warmly invite medicinal chemists, academic researchers, and anyone with a need for FEP to put Uni-FEP to the test. Simply open an issue in our GitHub repository using the provided issue template—upload the system you’re interested in (of course, only systems cleared for public sharing)—and we’ll run the calculation using Uni-FEP and return the results with you.🤝 We encourage everyone to participate the FEP Open Challenge. This initiative is not only about validating Uni-FEP but also about expanding the Apache 2.0-licensed benchmark dataset into a community-driven, high-quality, and large-scale resource.🥳 For more details on the FEP Open Challenge, check out our documentation（https://lnkd.in/g3ZffqXM) and join the discussion (https://lnkd.in/gU5RdgJK). #FEP #DrugDiscovery #ComputationalChemistry #Benchmarking #OpenSource

36

What happens when you connect state-of-the-art research with a real-world application?     Yesterday at ECML-PKDD 2025 in Porto, we presented GraphPCBs during the poster session.   Our project focuses on applying GNNs to optimize circuits in the schematics design phase, and it is a collaboration between CELUS and the University of Kassel, with the support of the BMFTR.   The response? Great interest, valuable discussions, and lots of curiosity from both the ML community as well as from people working with electronics.   Have you seen other examples where AI is making a difference in engineering workflows? 👇   #GraphPCBs #ECMLPKDD2025 #Electronics #MachineLearning #PCBDesign #CELUS

48

Portuguese-speaking coders: Want to stand out at work? Oi, galera! You don't need to overwork to stand out. Paulo Cardoso reacted to my post with 10 tips to stand out at work, sharing his thoughts on his YouTube channel. He translated my post and expanded on all 10 points. With the first 5 tips you can instantly stand out: 1. Deliver your work on time. 2. Document and automate your work. 3. Ask for the context behind your tasks. 4. Get interested in the business side of your work. 5. Speak up in meetings. For the other 5 and the more detailed explanation, watch Paulo's video here: https://lnkd.in/eSx88PhY

6

5 comentários

Credentials shouldn’t depend on PDFs. And verification shouldn’t take days. NOVA IMS is now issuing academic credentials as Verifiable Credentials, using blockchain technology aligned with W3C standards, OpenID, and the EBSI pilot. It’s the first production rollout of its kind in Portugal, placing NOVA IMS among Europe’s early adopters. Students can store credentials in their own digital wallet and share proof of a qualification in seconds with employers and institutions – who can verify authenticity automatically, reducing friction and fraud risk. The NOVA IMS Academic Credential Portal was developed by the Nova Blockchain Lab at NOVA IMS, led by Joana Rodrigues Pereira, with development by Francisco Ribeiro. NOVA IMS Dean, Miguel de Castro Neto, highlights its value for students and employers: “As a public school, we have a responsibility to lead solutions that bring real value to the country and to higher education. We developed this technology within NOVA IMS, aligned with European standards, to better serve our students and employers.” Full article: https://lnkd.in/dUpa-XaK #NOVAIMS #DataVisionaries #DataScience #ArtificialIntelligence #AI

39

You're starting a job in the software team at Shapeways. You'd expect to be diving into the code, right? Not so fast. When new engineers join Tiago São José's team at Shapeways, they don’t start by writing code. Instead, they spend a day in production — packing boxes, shipping orders and using the tools they’ll be developing. Because developing the best software comes from understanding how its used in the real world, and knowing the people who use it. ▶️ In this clip, Tiago explains why empathy is at the heart of engineering. See the full video on our blog: https://lnkd.in/e_QZQKC8

40

3 comentários