Civic Technology

YPS AI Agent

Policy Advocacy

Youth, Peace & Security National Youth Consensus

Social Entrepreneurship

UN Association of Afghanistan Kabul Model United Nations

Strengthening civic capacity through civic technology

A multi-layer AI agent grounded in a curated YPS knowledge base, making institutional knowledge and field research accessible while protecting privacy, preserving human agency, and keeping evidence at the center.

Developed for youth-led organizations, civil society, public institutions, UN and regional teams, and researchers working on YPS.

Explore YPS AI Agent See the architecture of the YPS AI Agent

What the YPS AI Agent is built to do

01

Identity-free access

Use the agent without creating an account or sharing personal identifiers.

02

Non-profiling

Conversations do not become personal or behavioral profiles.

03

Contextual memory

Keeps the thread of a session without turning context into identity.

04

Domain-specific intelligence

Grounded in Youth, Peace & Security knowledge, sources, and terminology.

05

Guardrailed functionality

Focused capabilities and clear boundaries reduce the space for misuse.

06

Multilingual access

Ask questions and explore evidence in the language that works for you.

07

Accessible interaction

A low-friction experience designed for different abilities and devices.

08

Human agency preserved

Supports human judgment and participation—never replaces them.

One agent, eight layers.

We have a single-agent system with a layered instruction stack and one ordered process. This keeps failures traceable and avoids the hidden handoff errors and added latency that can come with multi-agent orchestration.

Yet reliable reasoning depends on reliable evidence. The retrieval layer draws from nearly 500 public YPS documents across UN and regional frameworks, national strategies, academic research, and civil society field evidence.

So the system has a curated knowledge base that brings together institutional knowledge, research, and lived experience. Its categorized knowledge base helps the system balance different forms of evidence in its responses—distinguishing policy, recommendations, interpretation, and what reflects realities on the ground.

  1. 01
    Security

    Fires the moment a prompt arrives. Screens the prompt and decides whether to move forward or not.

  2. 02
    Scope

    Determines whether the question sits inside the YPS field, adjacent to it, or outside.

  3. 03
    Retrieval

    Draws relevant sources from the knowledge base only.

  4. 04
    Synthesis

    Combines the retrieved sources into a single evidence base for the response.

  5. 05
    Clarification

    Halts and returns a question to the user where the prompt is genuinely ambiguous.

  6. 06
    Drafting

    Structures the response, direct answer first.

  7. 07
    Citation check

    Verifies that the claims in the draft are actually supported by the sources cited, then numbers them.

  8. 08
    Formatting

    Applies length, readability, and output language.

Inside the architecture of YPS AI Agent An exploration of the platform’s layered agent architecture, evidence system, privacy model, and built-in safeguards. Read the publication
03B / Out of field Reject request Skip retrieval · return a generic scope response

User prompt selected.

What was taken out.

Several capabilities were deliberately narrowed to keep answers reliable, reduce the risk of misuse, and control cost.

Each of these is a trade made on purpose. A narrower tool is easier to hold to a standard. Every answer traces back to the curated corpus, and its boundaries stay visible to the person relying on it.

  1. 01
    Internet searchReliability

    The agent does not browse the web for information or build its own knowledge base from web content. Every response is grounded in the curated corpus.

  2. 02
    Document uploadMisuse · Cost

    Users cannot upload their own documents to the tool. This is a deliberate design choice to protect privacy and keep operating costs manageable so the tool can remain free.

  3. 03
    Grammar and text editingMisuse · Purpose

    The agent is not a document processor or a general-purpose assistant.

  4. 04
    Accounts and profilesPrivacy

    No login, cookies, or personal identification are required. The system does not request, collect, or store information that identifies users.

  5. 05
    Conversation retentionPrivacy · Cost

    Conversations are cleared when the browser is closed or refreshed. Contextual memory exists only within the active session, preserving continuity without compromising user privacy.

The question we started with

What would civic technology for YPS look like if privacy and knowledge-sharing were its first principles, and strengthening civic capacity its purpose?

Explore YPS AI Agent

The people behind it.

The agent was built by a small team independently. Product design, AI development and deployment, and advisory input were intentionally held as distinct roles, bringing technical, field, and institutional perspectives into the work.

Advisers were included because the quality of the system depends not only on how it is built, but also on how well it reflects the field it serves and the standards of evidence, reliability, and responsible use expected within it.