Background

Afghanistan is highly vulnerable to intense and recurring natural hazards that further risk growth and stability. Since 2000, natural disasters (i.e., droughts, earthquakes, epidemics, extreme temperature, floods, landslides, storms) have affected close to 19 million people, resulting in 10,656 deaths and US$173.11 million in total damages. Of these hazards, droughts have the most widespread impact and affect a larger population. The 5 significant drought events, occurring in 2000, 2006, 2008, 2011/12, and 2017/18, have affected over 17 million people with estimated damages totaling US$142.05 million. With its diverse topography, isolation of many vulnerable communities, and limited coping mechanisms, hazard events in Afghanistan, regardless of security factors, are ever more likely to turn into disasters with significant humanitarian and economic consequences.

The World Bank, through the Improving Livelihood Resilience to Climate Change in Afghanistan PASA (P500817), aims to identify investment opportunities for improving livelihoods through more climate-resilient agriculture and water sectors, including by enhancing proactive decision making to effectively mitigate the adverse impacts of natural hazards on life, livelihoods, and property.

Within this PASA, the World Bank, with the technical support of a consulting firm, has recently developed the Afghanistan Drought Decision Support Platform (AF-DDSP). This first of its kind innovative tool brings drought monitoring & seasonal forecasting for Afghanistan, utilizing high resolution remote sensing data to enhance predictive capabilities. The AF-DDSP aims to enable the Bank and other stakeholders to better monitor and forecast the drought conditions in the country and inform the design and prioritization of future operations.

The AF-DDSP system is built on a modular server-based architecture designed to support efficient data processing, storage, and access. This includes a Data Processing Server (dedicated to processing Earth Observation data and generating drought bulletins and analytical outputs), a Portal Server (which hosts all web services and APIs that provide access to drought related data through a web-based interface), a Database server (that manages and stores structured data, including geospatial and statistical datasets used by the platform) and a File Server (which handles the storage and retrieval of large files, such as EO data outputs, reports, and other binary assets).

As a proof-of-concept, the World Bank seeks to engage a qualified firm to develop and deploy an Artificial Intelligence (AI) conversational agent to enhance the AF-DDSP data accessibility and hence facilitate user experience.

Objective

The objective of this assignment is to develop an AI-powered conversational agent, integrated with the AF-DDSP, to enable users to query drought-related information using natural language. The solution will be built using a Retrieval-Augmented Generation (RAG) approach, leveraging an existing Large Language Model (LLM) to provide accurate, context-aware responses grounded in 25+ years of AF-DDSP data.

This system aims to enhance accessibility and usability of drought forecasts, historical indices, and analytical insights for both technical and non-technical stakeholders, eliminating the need for specialized database skills.

Specific objectives of the assignment:

• Design and implement a RAG-based AI system that connects a pre-trained LLM with the AF-DDSP database.
• Prepare and index AF-DDSP data, including drought indices, forecasts, and historical patterns, to support semantic search and accurate retrieval.
• Develop tailored prompt workflows and question-answer formats for different query types (factual, comparative, and predictive).
• Evaluate model performance using defined metrics such as factual accuracy, relevance, and latency.
• Deploy the conversational agent (e.g., chatbot) fully integrated with the AF-DDSP website for real-time interaction.
• Monitor system performance and gather user feedback to guide continuous improvement.

The objectives detailed above require strong coordination and working closely with the WB Task Team Leader (TTL) and the firm that has developed the AF-DDSP. 

SELECTION OF CONSULTING FIRMS BY THE WORLD BANK GROUP

REQUEST FOR EXPRESSION OF INTEREST (EOI)

Electronic Submissions through WBGeProcure RFx Now

ASSIGNMENT OVERVIEW

Assignment Title: DroughtWise – AI-Powered Language Model for Drought Forecasting and Decision Support

Assignment Countries:

• Afghanistan

ASSIGNMENT DESCRIPTION

Background

Afghanistanis highly vulnerable to intense and recurring natural hazards that further riskgrowth and stability. Since 2000, natural disasters (i.e., droughts, earthquakes,epidemics, extreme temperature, floods, landslides, storms) have affected closeto 19 million people, resulting in 10,656 deaths and US$173.11 million in totaldamages. Of these hazards, droughts have the most widespread impact and affecta larger population. The 5 significant drought events, occurring in 2000, 2006,2008, 2011/12, and 2017/18, have affected over 17 million people with estimateddamages totaling US$142.05 million. With its diverse topography, isolation ofmany vulnerable communities, and limited coping mechanisms, hazard events inAfghanistan, regardless of security factors, are ever more likely to turn intodisasters with significant humanitarian and economic consequences.

TheWorld Bank, through the Improving Livelihood Resilience to Climate Changein Afghanistan PASA (P500817), aims to identifyinvestment opportunities for improving livelihoods through moreclimate-resilient agriculture and water sectors, including by enhancingproactive decision making to effectively mitigate the adverse impacts ofnatural hazards on life, livelihoods, and property.

Withinthis PASA, the World Bank, with the technical support of a consulting firm, hasrecently developed the Afghanistan Drought Decision Support Platform(AF-DDSP). This first of its kind innovative tool brings drought monitoring& seasonal forecasting for Afghanistan, utilizing high resolution remotesensing data to enhance predictive capabilities. The AF-DDSP aims to enable theBank and other stakeholders to better monitor and forecast the droughtconditions in the country and inform the design and prioritization of futureoperations.

The AF-DDSPsystem is built on a modular server-based architecture designed to supportefficient data processing, storage, and access. This includes a Data ProcessingServer (dedicated to processing Earth Observation data and generatingdrought bulletins and analytical outputs), a Portal Server (which hosts all webservices and APIs that provide access to drought related data througha web-based interface), a Database server (that manages and stores structureddata, including geospatial and statistical datasets used by the platform) and aFile Server (which handles the storage and retrieval of large files, such as EOdata outputs, reports, and other binary assets).

As aproof-of-concept, the World Bank seeks to engage a qualified firm to developand deploy an Artificial Intelligence (AI) conversational agent to enhance theAF-DDSP data accessibility and hence facilitate user experience.

Objective

Theobjective of this assignment is to develop an AI-powered conversational agent,integrated with the AF-DDSP, to enable users to query drought-relatedinformation using natural language. The solution will be built using aRetrieval-Augmented Generation (RAG) approach, leveraging an existing LargeLanguage Model (LLM) to provide accurate, context-aware responses grounded in25+ years of AF-DDSP data.

Thissystem aims to enhance accessibility and usability of drought forecasts,historical indices, and analytical insights for both technical andnon-technical stakeholders, eliminating the need for specialized databaseskills.

Specificobjectives of the assignment:

• Design and implement a RAG-based AI system thatconnects a pre-trained LLM with the AF-DDSP database.
• Prepare and index AF-DDSP data, includingdrought indices, forecasts, and historical patterns, to support semantic searchand accurate retrieval.
• Develop tailored prompt workflows andquestion-answer formats for different query types (factual, comparative, andpredictive).
• Evaluate model performance using defined metricssuch as factual accuracy, relevance, and latency.
• Deploy the conversational agent (e.g., chatbot)fully integrated with the AF-DDSP website for real-time interaction.
• Monitor system performance and gather userfeedback to guide continuous improvement.

Theobjectives detailed above require strong coordination and working closely withthe WB Task Team Leader (TTL) and the firm that has developed the AF-DDSP.

FUNDING SOURCE

The World Bank Group intends to finance the assignment / services described below under the following:

• BB: Bank Budget
• TF0C8234: CREWS

ELIGIBILITY

Eligibility restrictions apply:

• [Please type list of restrictions]

SUBMISSION REQUIREMENTS

The World Bank Group invites eligible firms to indicate their interest in providing the services. Interested firms must provide information indicating that they are qualified to perform the services (brochures, description of similar assignments, experience in similar conditions, availability of appropriate skills among staff, etc. for firms; CV and cover letter for individuals). Please note that the total size of all attachments should be less than 5MB. Firms may associate to enhance their qualifications unless otherwise stated in the solicitation documents. Where a group of firms associate to submit an EOI, they must indicate which is the lead firm. If shortlisted, the firm identified in the EOI as the lead firm will be invited to the request for proposal (RFP) phase.

Expressions of Interest should be submitted, in English, electronically through WBGeProcure RFx Now

NOTES

Following this invitation for EOI, a shortlist of qualified firms will be formally invited to submit proposals. Shortlisting and selection will be subject to the availability of funding.

Only those firms which have been shortlisted will be invited to participate in the RFP phase. No notification or debrief will be provided to firms which have not been shortlisted.

If you encounter technical difficulties while uploading documents, please send an e-mail to the Help Desk at corporateprocurement@worldbank.org prior to the submission deadline.