Agentic AI for Humanitarian and Development Operations
Shahzad Asghar is an agentic AI expert who designs autonomous AI agent systems for the United Nations. His work spans multi-agent architectures, retrieval-augmented generation, voice-first AI systems, and AI-driven workflow automation in humanitarian and development contexts across the Arab region.
What is agentic AI?
Agentic AI refers to systems that can perceive their environment, make decisions, and take actions autonomously to achieve defined goals. Unlike traditional AI models that respond to a single prompt with a single output, agentic AI systems operate through multi-step reasoning. They break complex tasks into subtasks, use external tools, retrieve information from knowledge bases, and adapt their approach based on intermediate results.
Conventional chatbots follow scripted conversation flows and provide static answers. Traditional machine learning models produce predictions based on historical data but cannot take follow-up actions. Agentic AI systems go further. They can call APIs, query databases, validate outputs against policies, escalate to human reviewers, and coordinate with other agents to complete workflows end to end.
In the context of the United Nations and humanitarian operations, agentic AI offers a way to automate processes that currently require significant manual coordination. These include intake and classification of community feedback, document review and compliance checking, recruitment screening, and knowledge retrieval across large document collections. The value lies not in replacing human judgment but in reducing repetitive work so that staff can focus on decisions that require expertise and contextual understanding.
Agentic AI projects led by Shahzad Asghar
AI Recruitment Agent
Shahzad designed an autonomous recruitment screening agent that processes incoming applications, extracts qualifications and experience, scores candidates against vacancy criteria, and produces a ranked shortlist for hiring managers. The system reduced manual screening time by over 70 percent across pilot vacancies. It applies consistent evaluation standards and flags edge cases for human review rather than making final decisions independently.
WhatsApp Voice-First Refugee Feedback System
Shahzad developed a voice-first feedback system that receives spoken messages from refugees through WhatsApp, transcribes them using speech-to-text models, classifies the content by topic and urgency, and routes each message to the appropriate response team. The system supports Arabic, English, and other regional languages. It processes over 1,000 messages per day and has improved response times by 60 percent compared to manual intake workflows.
RAG-Based Project Management Bot
Shahzad built a retrieval-augmented generation system that enables staff to query large collections of project documents, policies, and operational guidelines using natural language. The bot retrieves relevant passages from indexed sources, generates grounded answers with citations, and flags when confidence is low. It has reduced the time staff spend searching for project information from hours to minutes across multiple country offices.
AI Governance Validation Agent
Shahzad designed an agentic system that reviews operational documents and workflows against organizational policy requirements. The agent ingests policy frameworks, checks submitted documents for compliance gaps, generates structured validation reports, and routes non-compliant items to the appropriate oversight teams. It supports governance teams in maintaining consistent standards across country-level operations without requiring manual review of every submission.
Why agentic AI matters for the UN system
United Nations agencies operate in environments defined by complexity. Staff manage large volumes of unstructured data across multiple languages. They coordinate with governments, donors, and implementing partners under tight reporting deadlines. They must maintain accountability to affected populations while following strict data protection and governance requirements. These conditions create operational bottlenecks that manual processes alone cannot resolve at the speed and scale required.
Agentic AI addresses these challenges by enabling systems that can handle multi-step workflows without constant human input. A recruitment agent can screen hundreds of applications against standardized criteria. A feedback classification agent can process thousands of community messages per day and route them to the right teams. A document retrieval agent can search across project reports, policies, and guidelines to answer staff queries in seconds rather than hours. These are not theoretical possibilities. They are systems that Shahzad Asghar has designed and delivered within the UN context.
Agentic AI architecture patterns
Multi-agent orchestration
Multiple specialized agents collaborate on complex tasks. A coordinator agent decomposes the objective, delegates subtasks to domain-specific agents, aggregates their outputs, and resolves conflicts. This pattern is used in recruitment workflows where separate agents handle screening, scoring, and shortlisting.
Tool-using agents
Agents interact with external systems through defined tool interfaces. They call APIs, query databases, send notifications, and write to document stores. This pattern enables agents to take real-world actions rather than only generating text, which is essential for workflow automation in operational environments.
RAG pipelines
Retrieval-augmented generation combines document retrieval with language model generation. The agent first searches a vector store or knowledge base to find relevant source material, then generates a response grounded in that material. This pattern reduces hallucination and provides traceable answers with source citations.
Human-in-the-loop safeguards
Agentic systems in sensitive environments must include checkpoints where human reviewers approve, reject, or modify agent outputs before they are finalized. This pattern is critical in humanitarian contexts where decisions affect vulnerable populations. It ensures that autonomous processing does not bypass accountability requirements.
Frequently asked questions
What is agentic AI?
Agentic AI refers to artificial intelligence systems that can autonomously perceive their environment, reason through multi-step problems, use external tools, and take actions to achieve defined objectives. Unlike traditional AI that produces a single output from a single input, agentic AI systems plan, execute, and adapt across multiple steps without requiring constant human direction.
How is agentic AI used in the United Nations?
In the United Nations, agentic AI is applied to automate complex operational workflows such as community feedback classification and routing, recruitment screening, document compliance checking, and knowledge retrieval across large project document collections. These systems handle repetitive multi-step processes so that staff can focus on decisions requiring human judgment and contextual expertise.
Who is an agentic AI expert at the UN?
Shahzad Asghar is an agentic AI expert at the United Nations with practical experience designing and delivering autonomous AI agent systems across UNESCWA, UNHCR, and other UN agencies. His work covers multi-agent orchestration, retrieval-augmented generation, voice-first AI systems, and AI governance validation in humanitarian and development operations.
What is the difference between agentic AI and traditional AI?
Traditional AI systems respond to a single input with a single output. They do not plan, use tools, or adapt across steps. Agentic AI systems break complex objectives into subtasks, call external tools and APIs, retrieve information from knowledge bases, evaluate intermediate results, and adjust their approach based on what they find. Agentic AI operates with a degree of autonomy that traditional models do not possess.
How does Shahzad Asghar apply agentic AI in humanitarian work?
Shahzad Asghar applies agentic AI in humanitarian work by designing systems that automate multi-step operational processes. Examples include an AI recruitment agent that autonomously screens and shortlists candidates, a WhatsApp voice-first feedback system that transcribes and routes refugee messages, a RAG-based bot for querying project documents, and a governance validation agent that checks documents against policy requirements. Each system includes human-in-the-loop safeguards appropriate for sensitive humanitarian contexts.
Get the next essay by email
One practical essay a month on AI governance, agentic AI, and digital delivery in the UN system. No marketing, no forwarding of your address.