An interesting engineering problem:
What happens when the user writes something we've never seen before?
We're building Edorena so that unknown language becomes a knowledge gap, not a system failure.
Real users will eventually tell us what to add.
#buildinpublic#AI#edorena
We don't want Edorena to tell users:
"This is a good decision."
That's not the job.
The system should say:
"This decision matches one of your rules."
The user decides what that means.
That distinction matters.
#ProductDesign#AI#edorena
We added confidence scoring to the rule engine.
Because:
exact match
and
semantic relationship
shouldn't necessarily have the same strength.
The engine should know not only what matched, but how strongly it matched.
#AI#NLP#buildinpublic#edorena
Thing we're learning while building Edorena:
The best systems have complicated reasoning underneath very simple interfaces.
user sees:
"Does this align with my rule?"
Underneath:
Parser → Graphs → Concepts → Evidence → Matching → Confidence
#BuildInPublic#AI#edorena
We're deliberately not asking an LLM to solve every problem inside Edorena.
Some problems are better solved with deterministic systems.
Graphs.
Rules.
Evidence.
Scoring.
AI where it helps.
Engineering where it makes more sense.
#AIEngineering#SaaS#founder#edorena
We built an Action Graph.
Not glamorous.
No fancy UI.
Just relationships like:
run → exercise
jog → exercise
workout → exercise
But this tiny graph changes how natural language rules behave.
#BuildInPublic#AIEngineering#edorena
We upgraded Edorena's rule model.
The interesting part isn't the code.
It's the shift:
"Does this sentence contain the same words?"
to:
Does this decision align with what the user said matters?
That's the product we're trying to build.
#buildinpublic#Productivity#edorena
Hardest parts of building Edorena wasn't writing the matcher.
It was deciding what a "match" actually means.
Exact match?
Related action?
Related object?
Concept match?
Context match?
We finally have a model for answering those questions.
#AI#NLP#buildinpublic#edorena
A rule shouldn't just be:
"Read books"
It should carry meaning.
What action?
What object?
What concepts?
What context?
How strong is the match?
That's the direction we're taking Edorena.
#AIEngineering#buildinpublic#edorena
We changed the way Edorena thinks about rules.
Before, a rule was basically something to match.
Now we're treating a rule as a structured model:
intent → action → object → concepts → context → evidence
That architectural opens up a lot.
#buildinpublic#AI#SaaS#edorena
🚀 Milestone
The new Edorena Rule Engine is now live.
Months of work:
✓ Semantic parsing
✓ Action graph
✓ Object graph
✓ Concept matching
✓ Confidence scoring
✓ Rule evaluation
On to the next chapter.
#buildinpublic#AI#startup#SaaS#edorena