Making AI trustworthy, explainable, robust and affordable
That is our mission. SynaLinks is an AI engineering team: we make SynaLemma, which helps your teams and agents answer your business questions, showing the rule behind every number, and we author the open-source engines it runs on.
Companies have capable models, not a way to check the answers
In two years we've worked with a dozen companies, and it was the same story every time: capable teams, valuable data, ambitious AI plans, and pilots that never shipped. The models were good enough. What was missing was a way to tell whether their answers were right, and where they came from.
We built SynaLemma for that: an AI that writes the rule behind every figure, has it verified by a logic engine, and runs it on the company's data, where it lives.
We built every layer of it ourselves
Most AI products combine someone else's model with someone else's framework. We wrote the framework the agent runs on, the logic engine that verifies its rules, the format the knowledge is kept in, and the product that assembles them. Three of the four are open source.
Because we wrote all four, the parts are designed to fit together, and we can tell you exactly what SynaLemma does and does not do. Synalinks, the framework, was developed under the supervision of François Chollet, the author of Keras.
Six principles
One definition per term
Each business term is defined once, over your tables, and every person and agent uses it.
The engine reasons, not the model
Every inference is made by a logic engine and can be replayed.
Verified before it answers
Every rule is proved well-formed, safe and terminating, then run on your data, before an answer is given.
Same rule, same SQL
The engine compiles the SQL from the rule. The model never writes it, so a definition gives the same query every time.
Every change on record
Your tables never changed, roles that limit people and agents alike, and every change a git commit with its author.
Sovereign by construction, not by contract
Your servers, the model you choose, and rules that remain your files in git, on open-source engines. They keep running if you stop working with us.