HARBORFIELDNeurosymbolic AI for engineering
Harborfield analyzes supported process-engineering XML documents, including P&IDs, PFDs, and block diagrams. PortLog combines a conversational interface with explicit engineering logic, keeping every answer tied to the rules, objects, and paths that made it true.
The product / four connected parts
Each part is useful alone. Together, they turn diagrams, requirements, and engineering judgment into a reviewable knowledge system.
Ask about a system the way you would ask an experienced engineer. Every claim stays attached to its evidence.
“Can P-101 still reach V-204 if valve V-14 is closed?”
No. V-14 is the only traversable connection between the pump discharge and V-204. Closing it removes the path.
Package standards, checks, and team knowledge as readable rules. See the exact logic before it runs, then reuse it across reviews.
Keep diagrams, conversations, and the rulepacks that govern them together. Context carries forward without hiding where it came from.
Run deterministic graph and Datalog reasoning beneath the conversation. Conclusions are reproducible, inspectable, and bounded by known facts.
reachable(From, To) :-
connected(From, Next),
open_path(Next),
reachable(Next, To).
answer(X) :- reachable(pump_101, X).Inside the logic engine
The model handles language. The engine handles entailment. PortLog keeps the two responsibilities separate, then brings their results back together.
Turn the engineer's intent into a bounded query over known objects.
Evaluate graph paths and rules deterministically against project facts.
Answer in plain language with the facts and rule trace attached.
Language models are excellent at interpreting intent and explaining complex ideas. They should not be asked to silently invent the engineering facts underneath an answer.
Symbolic reasoning gives those facts structure: explicit relationships, executable rules, and conclusions that can be checked again. The combination makes AI more useful without asking engineers to lower the standard of proof.
natural language + explicit rules + project evidence = grounded decisions
Start with a question