48 hours and 253 datasheets: how MAKE IT won the SpaceX Super Intelligence Digital Transformation Lab

Picture an engineer choosing a component for a new circuit board. The specs are buried in a datasheet PDF with dozens of pages, and the answer to one question, such as how a part behaves at a certain temperature, sits in a graph that has to be read by eye, sometimes with a ruler. Multiply that by every candidate part on the bill of materials and you have a slow, error-prone job that engineers repeat every day.

A team from Portugal decided to fix it in a weekend. MAKE IT, a product R&D startup incubated at UA Incubator in Ílhavo that develops technology products end to end: from electronics to the AI layer, and from idea and prototype to industrialization, won the SpaceX Super Intelligence Digital Transformation Lab on TAIKAI with Virtual Datasheet, a tool that turns datasheet PDFs into searchable, comparable data.

In 48 hours, the team built the application, the search engine behind it and processed 253 datasheets. Virtual Datasheet returns that value in seconds, linked to the source page.

A hackathon for Portuguese SMEs

The SpaceX Super Intelligence Digital Transformation Lab (SpaceXSI) took place on 26 and 27 October at Fórum Braga, supported by IAPMEI, Startup Portugal, and Invest Braga, and hosted on TAIKAI. The goal was to bring small and medium-sized companies and AI builders together to solve real digital transformation challenges in two days.

The response was strong: 218 participants registered and 62 projects were submitted. The first prize included USD 2,000 in AI credits.

The problem: datasheets are made for people, not for software

Every electronic component comes with a datasheet, a PDF that describes what the part does and how it behaves. Each manufacturer writes it in its own way, with its own names for the same parameters, and plenty of the information lives in charts and tables.

That works when a person reads one document. It breaks down when an engineer has to compare dozens of parts, or when a component runs out of stock and a replacement is needed fast. AI agents face the same wall: as Francisco Mendes, CEO of MAKE IT, explains, agents can read datasheets, but those PDFs were designed for human eyes, so the data stays locked inside.

Virtual Datasheet: point at a graph and read its curves as numbers, with the datasheet as source of truth on the left, all in sync.

What they built in 48 hours

Virtual Datasheet takes the PDF and turns it into structured data that can be searched. The UI shows the original document on the left and the extracted data on the right, and every value links back to the page it came from, so engineers can check the source in one click.

Behind the interface sits the engine, which does the heavy lifting:

• Vectorization and embeddings, so parts can be searched and compared by meaning, not only by exact keywords.
• Normalization, so the same parameter has the same name across manufacturers.
• Error detection, using the typical value ranges of each component family. It is in place for transistors, and the team is clear that it does not guarantee 100% accuracy, that's why they are improving validation algorithms.

On top of that, the tool can suggest alternative components when a part is out of stock, import your projects and bills of materials, and show parts and their neighbors in a three-axis visualization.

Virtual Datasheet: the front page promises 49 A, but without a heatsink the part carries about 8 A, so it is flagged.

"Is this a mockup?"

What convinced the jury was seeing it work. The demo was live and running on real data, and some people in the room asked whether it was a mockup. It was not.

Paul Dumaitre, Field Engineer at SpaceXSI, summed it up this way:

"This project is truly impressive. You upload a PDF with charts and technical information and the PDF comes to life."

Why open source, and the goal of one million datasheets

The 253 datasheets processed during the hackathon are a starting point. The team's goal is to reach one million. They are also exploring opening parts of the project to the community, so that the electronics community and AI agents could build on a common standard. The team will announce how they will continue the project before the end of the year. 

The team behind the project is led by its two co-founders, Francisco Mendes (CEO) and Pedro Magalhães (CTO).

What companies can take from a hackathon

For MAKE IT, the hackathon worked as a forcing function. Two days is not much time, so the team had to decide what mattered, build it and show it working. Preparation made the difference: the idea and the problem were clear before the event started, which left the 48 hours for building. The mentors and the feedback on site helped sharpen the pitch.

For companies, the lesson is practical. A short, focused event with a clear challenge can turn an idea into a working prototype faster than months of internal planning, and it puts the team in front of people who can validate it.

Francisco Mendes and Pedro Magalhães, who have organized and taken part in hackathons for years, are clear about why they matter for SMEs and what it takes to get value from one:

"A hackathon is like the Hyperbolic Time Chamber of Dragon Ball. The normal rules of time and working days do not apply, and it only works if the team arrives prepared, with the right mindset to innovate."

What is next

MAKE IT now moves toward the open source release and the one million datasheet goal. Engineers and component manufacturers interested in early access, or in verifying their datasheets, can contact MAKE IT at make-it.tech. You can follow the project through the links below:

• Virtual Datasheet pitch
• Demo video
• MAKE IT on LinkedIn
• MAKE IT on Instagram
• MAKE IT on X
• MAKE IT on YouTube

Have a challenge your company wants to solve with AI? Explore the next hackathons on TAIKAI and bring your team.