We built an AI that does not guess.
Every chatbot you know predicts the next word. Synapse Net started when our founder saw why that will never be enough, and built a model that checks its reasoning before it speaks.
The paper behind every chatbot.
In 2017 a team at Google published Attention Is All You Need. It introduced the transformer, built on two inventions of theirs: scaled dot-product attention and multi-head attention. It was made to improve translation, and it is the architecture inside ChatGPT and almost every AI model you have heard of.
Google did not move first with it. OpenAI published GPT-1 in June 2018, four months before Google's own BERT. When ChatGPT arrived in November 2022, Google answered with Bard months later. The full story.
Prediction is not reasoning.
Our founder, Giuliano Maroulis, started writing code at the age of eight. Studying how these models work, Giuliano saw a ceiling in the mechanism itself. A transformer predicts the next word from probabilities. It does not run logic. When the prediction matches the logic, the answer looks reasoned. When it does not, nothing inside the model notices, and it will explain the mistake as confidently as a fact.
Bigger models and more data make those mistakes rarer. They cannot make them impossible, because the mechanism stays prediction. Why LLMs cannot reason.
“It is all prediction, not logical reasoning. Text can sound natural and still say things that are illogical, or that do not exist at all. It pretends to be logical, but it never is.”
Start from physics, not from text.
So Giuliano started somewhere else. In nature, light, water and every physical system settle into the state that takes the least energy. PRISM applies that law to reasoning: an answer that fits what the model knows has low energy, and a contradiction makes the energy rise.
On top sits a layer that watches the model's own thinking, the way a person can notice their own emotions. PRISM learns basic truths, tells true from false, and only then produces words.
Small, real and shown live.
The proof of concept shows the mechanism works. You can steer it, it contradicts illogical input instead of going along with it, and it shows the full spectrum of what happened while it reasoned and how it classified each step. It proves the mechanism, not fluent language at scale.
Reaching basic conversation on the hardware we have would take about seven years of training. On NVIDIA Blackwell, one training run takes about a day, so we can rebuild and test the architecture as many times as it takes. That is what the seed round pays for.
Then we had to protect it.
An AI you can trust is worthless if its data can be stolen, decrypted or seized. So the rest of the stack followed: the Synapse Protocol for encryption that survives quantum computers, VISUM to keep data inside its country, and Aegis to run it all where nothing can reach it.
Lorenzo Festa joined to turn the technology into a company: strategy, funding and go-to-market across Europe.
The people building it.
Giuliano Maroulis
CEO & Lead ArchitectArchitect of the Synapse stack. When GPT-3 launched he saw the problem as architectural and spent three years designing the stack from first principles. He built the PRISM prototype himself.
Lorenzo Festa
Chief Operating OfficerOperator and founder. He took a client from zero to over €1M in annual revenue. At Synapse he leads strategy, funding and go-to-market across Europe.
See it working.
We show the PRISM proof of concept live, and share the technical brief and data room with qualified investors and partners.
Nothing on this website is an offer to sell or a solicitation to buy securities.