DilSe

DilSe Pakistan

Inside our AI architecture

How science becomes guidance: the dual-engine framework

DilSe does not rely on off-the-shelf generative models that produce generic advice. Our architecture translates peer-reviewed research into an interactive, real-time guidance system divided into two specialised modules: The Listener and The Partner.

How a message moves through DilSe

  1. Stage 01

    DilSe AI Core Engine

    One retrained conversational core, tuned on clinical methodology rather than generic advice patterns.

  2. Stage 02

    The Listener and The Partner

    Two specialised modules: reflective clarity when you need to understand the situation, interactive rehearsal when you need to practise saying it.

  3. Stage 03

    Cultural and clinical filters

    Extended family context, Culturally Adapted CBT and Affect Valuation Theory shape every response before you see it.

  4. Stage 04

    100% on-device privacy layer

    Local processing with end-to-end encryption and zero external server storage.

Module 1 — The Listener

Reflective deconstruction: understanding the situation before deciding what to say about it.

  1. Fact-assumption separationIncorporating Dr. Gurprit Ganda's Culturally Adapted CBT to isolate what happened from fear-based projections.
  2. Extended family mappingUsing Dr. Saunia Ahmad's frameworks to evaluate in-law dynamics and household hierarchies.
  3. Physiological and cognitive checkDrawing on Dr. Andrew Huberman's and Dr. Matthew Walker's research to evaluate physical fatigue and nervous system stress.
Sample Listener prompt

“You mentioned feeling deeply overlooked when your husband accepted his family's weekend plans without consulting you. Let's separate the facts from the assumptions: did he explicitly reject your preference, or did he fall back into an automatic joint-family decision pattern?”

Module 2 — The Partner

A rehearsal and roleplay engine for the exchange you are preparing for.

  1. Persona simulationSelect the relationship role you need to speak with, husband, mother-in-law or sister-in-law, with realistic cultural response patterns.
  2. Real-time tone and boundary feedbackInformed by Dr. Jeanne Tsai's research on emotional valuation, monitoring phrasing to avoid aggressive triggers within South Asian family contexts.
  3. Iterative refinementInspired by Lucas Miller's skill-acquisition techniques: pause, receive feedback and try the phrasing again.

Privacy-first architecture

  1. Zero external server storageConversations stay strictly on your personal device.
  2. No third-party AI data miningYour data is never used to train public language models.
  3. End-to-end local encryptionComplete data isolation on the device you are using.

Read the full detail of how conversations are handled in the DilSe privacy notice.

Keep reading

Try the two modules yourself

Start with the Listener when you need clarity, or the Partner when you already know what needs to be said.

Start a conversation