Nexus Impact bedrock AI

Humanitarian AI for all: 7 specialised agents, 11 languages, and offline-oriented access for research, emergency response, public health, grants, human rights, and women's health.

Comment

Nexus Impact Bedrock AI: Humanitarian AI for Online and Offline Access

Built by: Dr. Fetlework Workineh Asress
Location: Amhara, Ethiopia
Project focus: Amazon Bedrock integration, humanitarian technology, multilingual access, offline-oriented design, public health, and digital inclusion
Project stage: Early-stage prototype

Inspiration

I am Dr. Fetlework Workineh Asress, a public health researcher, humanitarian practitioner, and AI developer from Ethiopia's Amhara region. Through my experience in public health and humanitarian work, I have seen how conflict, displacement, unreliable internet access, language barriers, and limited resources can make it difficult for communities to obtain essential information and support.

Internally displaced people, refugees, host communities, women, young people, people with disabilities, and underserved communities can face several of these barriers simultaneously.

I wanted to explore how AI could help humanitarian workers, researchers, public health professionals, and community organisations access structured information and practical digital assistance, even in resource-constrained environments.

Developing this idea in Ethiopia also presented a personal technical challenge. Limited financial resources and difficulties accessing international payment services prevented me from verifying live Amazon Bedrock inference for this demonstration.

Rather than abandon the project, I continued building the application, implementing specialised agent workflows, developing an MCP server, and exploring locally available knowledge resources for selected offline functions.

Nexus Impact Bedrock AI grew from one central question: How can we make AI-powered humanitarian assistance more accessible when connectivity, language, and financial resources are limited?

What it does

Nexus Impact Bedrock AI is a mobile-friendly humanitarian technology prototype that brings seven specialised agent workflows into one application.

  • Humanitarian Emergency Response: Helps organise initial response priorities involving water, sanitation and hygiene (WASH), shelter, nutrition, and emergency planning.
  • Public Health and Epidemiology: Supports outbreak preparedness, surveillance planning, and public health programme development.
  • Research and Evidence Synthesis: Helps structure research questions, organise findings, and prepare evidence tables and literature-review materials.
  • Grant Proposal Design: Assists with developing project concepts, objectives, activities, and funding proposals.
  • Human Rights Documentation: Supports structured documentation of reported human rights concerns, with attention to privacy and safety.
  • Women's and Maternal Health: Provides health education and information about appropriate care-seeking.
  • Knowledge Base: Helps organise and reuse locally available knowledge and supported outputs.

The platform is intended for humanitarian practitioners, community organisations, researchers, public health professionals, and others working to address social challenges.

Rather than presenting AI as a replacement for human expertise, the project explores how specialised digital workflows can help people organise information and prepare structured outputs for further review.

The prototype is not a substitute for professional medical assessment, legal advice, verified emergency guidance, or established humanitarian procedures.

How we built it

I developed the application using React, TypeScript, Vite, Tailwind CSS, Node.js, Express, the Model Context Protocol (MCP), IndexedDB, the Web Crypto API, and Transformers.js.

The implementation brings together a browser-based interface, specialised agent workflows, locally available knowledge resources, and an MCP server.

The MCP server exposes specialised tools through a Streamable HTTP endpoint at /mcp. I also developed a browser-based Alexa+ simulator to demonstrate conversational workflows, voice interaction, and JSON-RPC requests and responses.

For the current demonstration, I implemented structured response templates and a bundled humanitarian knowledge pack. These provide a way to demonstrate the application's intended workflows without depending on a successful live cloud inference call. Simulated outputs are labelled SIMULATED.

The project includes an Amazon Bedrock integration in the codebase, but I have not yet verified a successful live inference call. I therefore do not claim that the current demonstration's responses were generated by a live Bedrock model.

The application also uses an installable Progressive Web App approach and locally available resources intended to support selected functions when an internet connection is unavailable.

Privacy-conscious features include personal-data redaction mechanisms, optional local encryption, and a Panic Wipe control. These features require further testing and security review before the application is used with sensitive real-world information.

Designed for online and offline access

Humanitarian needs do not disappear when internet connectivity becomes unreliable. I designed Nexus Impact Bedrock AI to explore how useful information and selected workflows can remain accessible under these conditions.

  • Online access: The hosted application and MCP server can be accessed when the required network connection and services are available.
  • Offline-oriented functionality: The installable Progressive Web App and bundled knowledge resources are intended to support selected local functions.
  • Low-connectivity environments: The design aims to reduce dependence on continuous internet access where local functionality is available.
  • Information sharing: Available features include QR-code sharing, Markdown export, and printing.

Offline support has limitations. Cloud inference, remote API calls, and other server-dependent features still require connectivity. Further device and network testing is needed to establish the full extent of offline functionality.

Multilingual accessibility and inclusion

Although Nexus Impact Bedrock AI was inspired by challenges I have encountered in Ethiopia, I want its design to be adaptable to other communities and countries.

The interface supports 11 languages:

English, Amharic, Tigrinya, Afaan Oromoo, Somali, Swahili, Hausa, Kinyarwanda, Arabic, French, and Portuguese.

Accessibility features include plain-language mode, large text, voice input, read-aloud, icon-first navigation, and right-to-left layout for Arabic.

These features are intended to reduce barriers to using digital tools. However, language support alone does not guarantee accurate communication or cultural relevance. Native-speaker review, community participation, and usability testing remain important next steps.

Challenges we ran into

One of my biggest challenges was developing the project with limited financial resources. As a developer based in Ethiopia, I faced difficulties accessing international payment services to fund and test live Amazon Bedrock inference.

Although the Bedrock integration is present in the code, I could not verify a successful live inference call for this demonstration. I chose not to describe simulated responses as live AI-generated results.

Instead, I implemented structured templates and locally available humanitarian knowledge resources so I could continue demonstrating the application's workflows. This allowed development to progress while making the limitations clear to judges and users.

Another challenge was designing for unreliable internet access. An application can be installable without every feature working offline, so I had to distinguish locally available functionality from features requiring remote services.

Developing multilingual and accessibility features also highlighted the need for careful testing. Translation quality, cultural relevance, privacy, and security cannot be established simply by implementing interface controls.

These challenges taught me to work within technical and financial constraints, make practical implementation choices, and document what has been built, what has been demonstrated, and what remains unverified.

Accomplishments that we're proud of

I am proud to have developed Nexus Impact Bedrock AI as an early-stage humanitarian technology prototype despite the financial and technical challenges I encountered.

My main accomplishments include:

  • Bringing seven specialised humanitarian and public-interest workflows into one application.
  • Implementing a multilingual interface supporting 11 languages.
  • Developing accessibility features for text, voice interaction, and navigation.
  • Building an offline-oriented Progressive Web App experience with bundled knowledge resources.
  • Developing an MCP server that exposes specialised agent tools through a Streamable HTTP endpoint.
  • Creating a browser-based simulator for demonstrating conversational workflows and JSON-RPC interactions.
  • Implementing structured simulated responses that allow the application to be demonstrated without claiming unverified live AI inference.
  • Incorporating privacy-conscious mechanisms for further security testing.
  • Maintaining an open-source codebase that other developers can inspect and potentially adapt.

I am especially proud that I continued developing the project rather than allowing limited access to paid cloud services to stop experimentation.

For me, the achievement is not that every technical challenge has been solved. It is that I have built a foundation that can be tested, improved, and evaluated transparently.

What we learned

Building Nexus Impact Bedrock AI taught me that responsible AI development requires more than connecting a model to an application.

First, I learned to distinguish between an implemented integration and a verified, functioning integration. Having Amazon Bedrock-related code in a repository does not establish that a live model call succeeds or that its output is reliable.

Second, I learned that offline functionality must be evaluated feature by feature. Locally available knowledge can support selected tasks, but it cannot automatically replace cloud inference or remote services.

Third, I learned that multilingual accessibility requires ongoing work with the communities who will use the application. Translations, cultural relevance, and usability must be evaluated rather than assumed.

Finally, I learned that humanitarian technology must prioritise safety, privacy, and evidence. Simulated demonstrations can help explain a product concept, but they do not establish factual accuracy, production readiness, or real-world impact.

These lessons will guide the next stage of development.

What's next for Nexus Impact Bedrock AI?

My goal is to move the project from an early-stage prototype toward a tested, reliable, and useful humanitarian support platform.

My next priorities are:

  1. Verify Amazon Bedrock inference: Obtain the necessary cloud access or credits, test successful live model calls, and document the model, configuration, errors, and observed results.
  2. Strengthen the agent architecture: Evaluate the specialised workflows and determine how verified model outputs can be integrated into appropriate tasks.
  3. Test offline functionality: Validate the installable application and bundled knowledge resources across real devices, browsers, and connectivity conditions.
  4. Improve multilingual quality: Work with native speakers and community organisations to review translations and cultural relevance across all 11 supported languages.
  5. Improve security and privacy: Test personal-data redaction, local encryption, and the Panic Wipe feature, and arrange a security review before processing sensitive cases.
  6. Evaluate with intended users: Engage humanitarian organisations, researchers, public health professionals, and community groups to identify needs and assess usability.
  7. Improve technical reliability: Expand automated testing, documentation, error handling, and reproducibility so developers and partners can evaluate the codebase.
  8. Measure results responsibly: Establish evaluation criteria for functionality, accessibility, response quality, and user experience before making claims about impact.

I also hope to collaborate with technical partners and humanitarian organisations that can help evaluate the project and support its responsible development.

Potential impact

Nexus Impact Bedrock AI aims to explore how AI-assisted tools and locally available knowledge can help reduce barriers to accessing structured information in humanitarian and public-interest settings.

Potential applications include emergency-response planning, public health preparedness, research support, human rights documentation, maternal health education, and grant development.

Its multilingual interface and offline-oriented design may be particularly relevant to communities and organisations facing connectivity, language, and resource constraints.

An open-source approach can also make the implementation available for inspection, experimentation, and potential adaptation by other developers and organisations.

The project remains a prototype, and these benefits have not yet been established through formal field evaluation. Demonstrating its value will require technical validation, user feedback, appropriate safeguards, and partnerships.

What I want to achieve

My vision is to develop humanitarian technology that can be adapted to different local contexts, supports access across language barriers, and remains useful for selected tasks when connectivity is limited.

As a public health researcher, humanitarian practitioner, and AI developer from Ethiopia, I want to connect practical community needs with responsible technical innovation.

Nexus Impact Bedrock AI represents my effort to build toward that goal with the resources available to me, while keeping the project's capabilities and limitations transparent.

Built from the realities of Ethiopia, Nexus Impact Bedrock AI aims to make digital assistance more accessible across languages, borders, and connectivity barriers.

Try the project

Built With

  • accessibility
  • amazon-bedrock
  • artificial-intelligence
  • express.js
  • human-rights
  • humanitarian-technology
  • indexeddb
  • model-context-protocol-(mcp)
  • multilingual
  • node.js
  • offline-first
  • open
  • progressive-web-app
  • public-health
  • react
  • tailwind-css
  • transformers.js
  • typescript
  • vite
  • web-crypto-api
Share this project:

Updates

Submission history