Ekam, a design method that runs with AI
I built Ekam over three months: a design method that derives a product's structure from evidence. AI does the derivation, and a person decides at every gate. Every step lands in a map and a knowledge graph, so any claim can be traced to its source.
This is solo work. I tested it on a real company's website as a two-day proof of concept, and structured this site with it. The agentic mortgage platform record shows the same AI-assisted research inside a team.
- Role
- Sole designer and builder
- Team
- None. Self-directed
- When
- July to September 2026
- Built as
- A Claude skill, with scripts that build the map and graph
A method, not a tool that designs for you
Ekam reads public evidence about a product: its pages, reviews, competitors and announcements. From that evidence it derives what the business needs and what its users need, finds where the two meet, and turns that into principles and a structure.
It stops there. From structure onward, it frames options and assesses each one against what it derived, and I choose. An assessment is information, never a verdict.
July and August went into groundwork. The method took shape in September, as a Claude skill with scripts of its own.
Every claim, traceable to its source
Each run produces two views of the same derivation, built by Ekam's own scripts. The map shows every stage as a card, with each segment, brief, principle and area it derived.
The knowledge graph links every claim to the evidence it came from. A reader can start at any decision and walk back to its source.
Built so the method cannot fool me
A method that runs with AI can make a conclusion look derived when it was only assumed. Ekam has four guardrails against that.
- A sealed predictionBefore any evidence is read, I write down what I expect. If the result only repeats it, the stages did no work.
- Labelled inferenceAnything the evidence does not show directly is marked as inferred.
- Gates for a personThe run stops at decisions. Options come with assessments, and the choice is mine.
- Checks on every changeScripts check the writing for plain, short sentences, and rebuild the map and graph.
Confident fit
The right intervention, understood in system context
From the proof of concept below: the prediction against what the run derived.
Two days on a real company's website
To test the method, I ran it on Niswey, a HubSpot partner that wanted a visual refresh of its homepage. It took two days, and nobody commissioned it.
The run read 61 pages sampled from 850, 55 ratings and reviews, and four competitor sites. None of the four let a buyer start from one exact problem and see its wider effect.
The method's own first principle then said to lead with the visitor's problem. I overrode it and kept the homepage offer-led, because visitors should understand the company before they understand how its site was designed. The run recorded why.
"The public page should feel simple because the underlying reasoning is sound, not because the reasoning has been copied into the page."
Built, in use, and still solo
Ekam is built and in use. This site was structured with it, and each of its decisions was assessed against principles before I chose.
The proof of concept left a gap I recorded. Its log said why I overrode the method, but not which principle I overrode. Later runs assess every decision against the principles before I choose.
Ekam is self-directed. Nobody could say no to it, and it has not yet been used inside a team. The Niswey prototype was sent to them and is not live, so there is no outcome to report.
Research and design system, agentic mortgage platform
Research run through an AI framework I wrote, and a design system 80 to 85% built.
Also: Payments bank B2B platform, Multi-location digital presence platform. Every role, team and date.