LumX
LiveConsumer AI serviceLive on the web since 2026

Live service case

somesignal

A live service where the connections you make accumulate as a map

The front page is an input rather than a login button: a birth date alone returns a reading that ties saju and MBTI together in one place, and the date entered is not stored. An invitation works the same way — opening the link shows a form instead of a sign-up wall, and filling it opens both people's result at once. Either result can leave as a vertical card drawn in the colours and characters of its five-element reading, which opens for whoever receives the link without an account of their own. Signing up is what keeps a relationship rather than what unlocks it: an account turns the reading into a star on a signal map, where plans, a relationship timeline, the map lighting up, and a connection's approaching birthday are the reasons to come back. Push delivery is switched off, so nothing arrives outside the app.

somesignal is live and operating, with its decisions and next improvement priorities documented here.

Visit somesignalLeave feedback

Service development record

How a compatibility reading turned into a relationship map

The product began as a reading two people could talk about, and operating it moved the centre of gravity from the reading to the map of connections it left behind.

  1. 01Observation & evidence

    A lightweight result that rarely remains in conversation

    Question examined
    How much can compatibility content help two people talk after the entertainment ends?
    Decision made
    Focus on specific prompts people could discuss together rather than an experience built around consuming a score.
    Evidence used for the decision
    The product observation was that compatibility content often stops at generic language and a single score while leaving cultural context out of the result.
  2. 02Problem & principle

    Relevance and emotional responsibility over precision

    Question examined
    How can a result feel personal without making claims about the relationship?
    Decision made
    Use structured personality inputs and cultural context, while refusing to present a score as objective truth or a verdict on the relationship.
    Evidence used for the decision
    The more specific AI-generated language appears, the more easily it can be treated as accurate beyond its actual trustworthiness or usefulness.
  3. 03Hypothesis & scope

    Test the first result and the reason to return

    Question examined
    What is the smallest product flow that can examine the hypothesis?
    Decision made
    Build input, AI interpretation, a compatibility result, and conversation prompts while excluding psychological diagnosis and authoritative relationship advice.
    Evidence used for the decision
    The initial learning scope centered on consumer AI positioning and whether value remained after the first generated result.
  4. 04Experience & system

    A consumer AI product built for fast validation

    Question examined
    How can a small experiment become a web service people can actually use?
    Decision made
    Connect input to generated results with React, serverless functions, Supabase PostgreSQL, and OpenRouter, then release it on an independent domain.
    Evidence used for the decision
    The stack minimized technical complexity while keeping the result context and language quick to adjust as the hypothesis evolved.
  5. 05Launch & learning

    The first result was not the product

    Question examined
    Is there a clear reason for someone to return after viewing the result?
    Decision made
    Split the two jobs: let the reading be the door, opened to anyone without an account, and keep the accumulating map as the reason to stay.
    Evidence used for the decision
    The verbs changed first — nothing is sent, connections are made — and then the loop shipped: plans, a relationship timeline, and being told about a connection's birthday. The login wall in front of the first result came down afterwards, and the cost ceiling it had been quietly enforcing was rebuilt before it did. Place recommendations, tapping a star, and the question cards were removed in the same pass, since each assumed the other person would take part. Push delivery stays switched off, so no claim is made about notifications, and no usage figures are published, so reach is not claimed either.
Signals used for the next decision
Current focus
Whether a reading seen without an account becomes a map worth keeping
Operating status
Live and operating
Next decision
Narrow the target definition beyond 'Korean users'

These are inputs for the next operating or validation decision, not outcomes already achieved.

Technology used to build it

  • React
  • TypeScript
  • Vercel Functions
  • Supabase PostgreSQL
  • OpenRouter

What operating it taught

  • Saju apps read saju and MBTI tests read MBTI. Tying the two together in a single sentence turned out to be the only real difference this product has, so the strategy moved to putting that sentence in front of people before anything else is asked of them.
  • A sign-up wall in front of a result was collecting a birth time and an account from people who had not yet seen anything worth giving them for. Moving the account to the moment someone wants to keep a relationship made it a way to save rather than a price of entry.
  • Removing a finished feature is a product decision, not a failure. Place recommendations needed constant curation and belonged to a different product than reading a relationship, so it was taken out rather than maintained.

What is still undecided

  • Find out whether a reading seen without an account converts into a map worth keeping
  • Narrow the target definition beyond 'Korean users' — the age and situation are still undecided

Change history

What changed in this service

Dated by when the change shipped in the product, not by when this page caught up. Lines are added rather than replaced, so what the service withdrew stays readable next to what it gained.

  1. Added

    Turned the front page from a login button into an input: a birth date alone now returns a reading that ties saju and MBTI together in one place, and the date entered is not stored.

  2. Withdrawn

    Removed tapping a star and the question cards. Both assumed the other person would take part, which no longer fits a service that shows a result before asking for anything.

  3. Added

    Stopped asking an invited person to sign up. Opening the link shows a form, and filling it opens both people's result on the spot — an account is asked for only when someone wants to keep the relationship on their map.

  4. Added

    Added a share card that draws the reading in the colours and characters of its five-element reading. Whoever opens the card link sees the same card without signing up.

  5. Withdrawn

    Removed signal spots, the offline places suggested for two people to meet. Making plans stayed; choosing the place did not — it needed constant curation and sat outside what this service is.

The question I want feedback on now

What would make a lightweight relationship product worth returning to after its first result?

Leave feedback