BACK TO INDEX
/01ACT 1R&D Lab · Empathetic AI2025

Lauren's Crib

An empathetic AI that reads how you behave, not just what you type.

The question

Can a machine care for you without surveilling you?

The problem / context

Advice tools tend to either flatter you or judge you, and the ones that try to help usually do it by hoovering up everything they can. Real empathy needs to read behaviour honestly - but that reading runs straight into the boundary of privacy. How much of a person should a model be allowed to know before help quietly becomes surveillance?

The outcome / reflection

An empathetic chatbot that works as a relationship advisor by inferring patterns from how people behave over time, not just the words in a single message - paired with a deliberate, designed line around what it stores and what it is allowed to surface. The harder the model leans on behaviour, the more explicit that boundary has to be.

// OVERVIEW

Lauren's Crib is the R&D lab where the rest of this work gets built: deep tech, the circular economy, and complex LLM workflow automation. Its flagship experiment is an empathetic AI chatbot that acts as a relationship advisor - a system designed to understand the person it talks to well enough to actually help, while treating that understanding as a responsibility rather than a feature.

// MAIN IDEA

Behaviour as the signal, privacy as the constraint

Most chat advice treats each message as the whole story. Lauren's Crib treats the message as one data point in a longer pattern of behaviour - tone over time, what gets avoided, what repeats. That makes the advice sharper, but it also means the system knows more than a user might expect. So the privacy boundary is not an afterthought bolted on at the end; it is the design constraint the whole thing is built around. The lab keeps re-drawing that line as the model gets better at reading people.

// DESIGN PROCESS

  1. Frame the relationship

    Define what 'advisor' means: not a yes-man, not a therapist, but a system that reflects patterns back to you honestly.

  2. Map the behaviour signals

    Decide which behavioural cues are worth reading and which are off-limits, before writing a line of prompt logic.

  3. Draw the privacy line

    Specify what is inferred in-session versus stored, and what the model is never allowed to volunteer.

// TECHNICAL DETAILS

  1. LLM workflow orchestration

    Chained prompts and tools that turn a stream of messages into a maintained model of the relationship, rather than one-shot replies.

  2. Behavioural inference

    Lightweight pattern extraction over conversation history to surface tendencies the user may not name themselves.

  3. Boundary enforcement

    Explicit guardrails on retention and disclosure so the empathetic reading never leaks past its consented use.

// MAKING

Implementation

Built inside the lab's LLM-automation stack: the advisor is a workflow, not a single prompt, so the privacy boundary can be enforced at each step rather than hoped for at the output.

// BACKGROUND

The lab grew out of a simple frustration: tools that claim to understand you rarely earn the right to. Lauren's Crib is the attempt to earn it.

// WHAT DID I LEARN?

  • Empathy and surveillance share the same raw material - behaviour - so the ethics have to be designed in, not reviewed afterward.

  • Advice that names a pattern lands harder than advice that answers a question.

  • A clear boundary makes a system more trustworthy, not less capable.

// CREDITS

  • Laurensius Rivian Pratama - concept, AI engineering, system design

// WHAT’S NEXT

Tighten the boundary tooling so the privacy line is legible to the user in real time, not just enforced behind the scenes.

NEXT / 02Babblin Wastebin