PRODUCT / AI / UX / 2026

Public product

Olli AI

From scattered intentions to daily action.

MY WORK Product concept, decision logic, interface design and implementation planning.

LIVE PROJECT

DESCRIBE YOUR SITUATION

Ask your own question in the app. Below it, a separate example shows how Olli can help break down a decision: “I want to start a project, but I am not sure I can handle the scope.”

Try it here: tap the buttons and scroll inside the phone.

Olli AILOADING…
Open in a new tab
01

Voice

A person can describe a situation in their own words without preparing a special prompt.

02

Free text

The interface asks for the situation, not for technical parameters.

03

Example requests

Not sure where to start? Choose one of the example questions.

ILLUSTRATIVE EXAMPLE04 / 04

Decision map

Two routes are compared: commit to a large scope immediately, or test one narrow scenario first.

First reversible stepBuild one scenario and record the actual time before committing to the full project.
01

What is known

In this illustrative example, two evenings a week are available for the project.

02

What matters

The current commitments need to remain stable while the idea is tested.

03

What is missing

The time required for a useful first version is still unknown.

04

What to try

Build one narrow scenario, record the actual time and revise the route if it does not fit.

The application above is interactive. The decision card below it is an explanatory example, not a record of a personal conversation.

MY CONTRIBUTION

Product concept, decision logic, interface design and implementation planning.

DELIVERED

Decision map, conversation flow and prototype interface.

APPLICATION

AI assistants, advisory systems and decision-support interfaces.

WHAT IS SHOWN

The live product is embedded above. The decision-map panel is an explanatory example.

SYSTEM VIEW

HOW IT WORKS

A person describes a situation in everyday language. The system separates facts from assumptions, connects them to values and criteria, compares options and returns a structured decision map instead of generic advice.

01

Dialogue

Open request and clarifying questions

02

Context analysis

Facts, values and unknowns

03

Decision model

Options, risks and criteria

04

Action map

Conclusion and reversible first step

The main steps in the system, with technical details left out.

FROM CASE TO YOUR BUSINESS

What you can use

An assistant that turns a complex question into a clear conversation.

The live product can be opened above; the decision card is an explanatory example.

HOW IT IS BUILT A web interface, an AI dialogue and a structured decision map; project data is organized in a database.

Ask in your own words

The assistant asks for missing context before comparing options.

See the alternatives

Facts, assumptions and possible steps are separated so a person can review them.

Keep the decision with you

The conversation helps clarify a next step without choosing for the person.

FOR YOUR PROJECT

For a business, this approach could become a guided consultation, product selector or assistant for recurring customer questions. The questions, data and boundaries would be designed for your task.

Discuss your task
Engineering details and safeguards
TECHNICAL OVERVIEW

FROM WORDS TO A SYSTEM THAT CAN REVISE ITSELF

Conversation, memory and the decision map are connected. When a person clarifies information, the system updates its conclusions and the next step.

01

Signal

The current message is weighted more strongly than old context. A router selects only the relevant decision, safety or voice capability.

02

Context

Conversation memory, confirmed meanings and evidence are assembled into a compact working context — not an unlimited personal dossier.

03

Decision

The model separates what is known from what is assumed, compares routes and produces a structured decision card instead of generic advice.

04

Reality

The person confirms, edits or rejects the interpretation. New evidence changes the map and makes the next answer more precise.

LIVING MEMORY

An interpretation needs confirmation

The map keeps the source, confidence and evidence count. Repeated evidence may strengthen an item, but the person’s explicit confirmation remains decisive.

01Observationone source
02Testingcautious hypothesis
03Repeatedindependent evidence
04Confirmedperson decides
CONTROL LAYER

Useful without pretending to know the person better than they do

01

Evidence before confidence

A new observation starts as a hypothesis. Confidence grows only when independent conversations support it.

02

Human retains control

The person can confirm, edit, resolve or reject an item. The system does not silently turn an inference into a fact.

03

Sensitive wording

Sensitive patterns are phrased as possibilities; diagnostic language is filtered from generated map items.

04

Urgent route

Immediate danger, medical urgency and self-harm signals bypass the ordinary answer and receive a dedicated safety response.

IMPLEMENTED FOUNDATION

Node.js / Express service · PostgreSQL state and event history · structured JSON schemas · prompt and safety routing · OpenAI with optional Gemini and Claude modes · browser and unit tests

This diagram explains the repository architecture; it is not a claim that every experimental module is enabled in production.
A DECISION I MADE

A structured decision map instead of one long recommendation.

Why

Facts, criteria, unknowns and the next step remain visible and can be revised separately.

Trade-off

The conversation needs extra clarification before it can produce a useful map.

FOR YOUR BUSINESS

An assistant for your users?

Tell me what people need help understanding and what a useful result would look like.

Discuss an AI assistant Agree on scope and a way to verify the result.