Building SIGNAL
SIGNAL AI
SIGNAL started with a deceptively simple question, what if the context buried in everyday customer conversations could become useful without turning every conversation into another form to complete? Building it meant deciding not only what AI could do, but what it should do, what evidence it needed, how uncertainty should appear, and where human judgement needed to remain in control
Context
Start with evidence
SIGNAL begins with what was actually observed: the conversation, its context, and relevant relationship history. Interpretation should be grounded in evidence, not presented as something the system simply "knows"
Uncertainty
Make interpretation visible
AI can identify patters and suggest what may be changing, but an inference is not a fact. SIGNAL keeps that distinction visible so people can understand what the system noticed and apply their own judgment.
Judgement
Know when to defer
Not every signal should become an action. SIGNAL distinguishes between what AI can surface, what it can recommend, and what should remain a human decision based on uncertainty and consequence.
From Principle to Product
Designing degrees of autonomy
The challenge wasn't simple deciding what AI could do. It was deciding how much authority the system should have at each point in the experience, based on the strength of the evidence, the level of uncertainty , and the consequence of getting it wrong.
SURFACE
Make the evidence useful
Show what was observed and the context behind it before asking the user to interpret what it means.
RECOMMEND
Help decide what comes next
Suggest a next step when the evidence supports it, while keeping the reasoning visible to the user.
ACT/DEFER
Match autonomy to consequence
Automate low-risk actions when confidence is sufficient; defer when uncertainty or consequence calls for human judgement.
HOW I BUILT IT
Prototype quickly. Operationalize deliberately.
I used AI-assisted prototyping tools to turn SIGNAL from an idea into an interactive product experience quickly, making it possible to explore workflows, test assumption, and refine how the system should behave before investing in production development. A production implementation would move through the methodology phases defined in my AI Product Framework, adding the constraints, evaluation, governance, and documentation needed to take an AI experience responsibly forward. The prototype proves the experience. The framework defines what it would take to operate it responsibly.
Explore the Prototype