New publication: Self-Improving Cost Estimation for Architectural Projects: A Feedback-Driven Approach
ReadCost estimating is a memory and reinforcement problem. Arkesif models it with neurobiological principles.
A brain does not freeze what it learns; it rewires itself with every new experience. Arkesif's learning layer works in the same order: it collects signals, strengthens what works, prunes what does not, and rebuilds its memory at regular intervals.
A price is not a number known from birth; it is shaped by experience. When a practice prices a new project it looks at past jobs, mistakes and corrections. A learning system has to do the same.
Each row holds three things: the biological principle, how it appears in the product, and the metric we measure. Real mechanics instead of a made-up 'neural network' story.
Every correction is a signal. The signal becomes a feature, the feature updates the model, the model produces a new suggestion — and the loop closes.
Signal
Feature
Model
Suggestion
Feedback
Being honest is part of the research: we separate what runs today from what is on the roadmap.
Running today
On the roadmap
A learning system is only trustworthy if it says what it does. Not a black box: every suggestion shows its source, sample count and confidence range. Nothing enters the estimate without your approval, and models run separately for each company.
The sources below are the scientific basis of the four mechanisms the engine actually uses. Each line states which mechanism it grounds.
Co-occurrence learning (Hebb rule)
Hierarchical shrinkage — Trends in Cognitive Sciences 10(7)
Recency weighting and consolidation — Science 367(6473)
Statistical learning of regularities — J. Exp. Psychology: General 134(4)
Sources are listed for attribution only; the text and diagrams on this page are original to this project.
The full research
Measurement method, dataset and publications.