Overview and Objectives

 AI is evolving at a rapid pace, with significant progress in language models as well as in speech and speech–language models. Alongside this, language models combined with tools have increasingly taken the form of agents that can perform real-world tasks. We worked on a few projects in a similar domain last year, and this year our focus is on exploring state-of-the-art tools that can be integrated directly into our mobile applications.

Our goal is to build a voice agent that can be embedded into our mobile apps to improve accessibility and inclusivity—especially for users who face barriers due to technology constraints or differences in native languages. This voice agent will allow users to interact with the app entirely through voice, perform certain actions on their behalf, and understand different features and use cases of the app without relying on traditional UI interactions.

The solution is intended to use a lightweight, small-footprint model with minimal memory requirements, making it suitable for mobile environments while still delivering effective voice-based interaction.

Description

At a high level, the core of this project will involve research and exploration into how AI-powered voice agents can be effectively integrated into mobile applications. A major focus will be identifying which features and capabilities make the most sense to expose through voice interaction, and what kinds of user operations can be reliably and safely performed using a voice-first interface. This includes understanding user experience considerations, accessibility requirements, multilingual support, and limitations specific to mobile platforms.

Another key aspect of the project will be evaluating the existing ecosystem of tools, frameworks, and models that can be used to build such voice agents. This includes researching speech-to-speech systems, lightweight language or speech–language models suitable for mobile environments, and agent frameworks that support tool usage. The intern will be expected to assess trade-offs between different deployment strategies, such as using hosted/cloud-based models versus running models locally on-device, taking into account factors like latency, privacy, cost, reliability, and memory footprint.

Skills

Kotlin / Mobile app development

Python

AI agents and tool-based workflows

Speech-to-Text (STT) systems

Text-to-Speech (TTS) systems

Voice-first UX / accessibility

On-device vs hosted model deployment

Audio streaming and processing

Resources

https://docs.ultravox.ai/overview
https://www.hume.ai/
https://www.liquid.ai/blog/lfm2-audio-an-end-to-end-audio-foundation-model

Mifos Repositories

https://github.com/openMF/mifos-mobile

GitHub - openMF/community-ai: This repo is a set of Mifos Community AI tools that support the Mifos products. Mifos X , Payment Hub EE, Mifos Gazelle. It is managed in Jira https://mifosforge.jira.com/jira/software/c/projects/AI/boards/100

Impact

Our mobile app is widely used in regions where English is not the primary language, and it also supports multiple Digital Public Goods (DPGs) and public infrastructure initiatives aligned with UNICEF’s goals, making inclusivity a core priority. By enabling seamless voice-based interaction, this project will allow users who may not speak English or who face technological barriers to use the app effectively through simple voice commands. In addition, the voice agent will help users better understand the app by providing guided, voice-only onboarding and explanations of its features. Overall, the project aims to significantly lower access barriers, improve user experience for diverse communities, and create meaningful social impact by making the app more accessible and inclusive.

Provisional Mentors

Aru Sharma, TBC

Channels to Discuss

Slack: https://mifos.slack.com

Channel: #mifos-mobile #wg-ai-for-all

Overview and Objectives

Currently Mifos Gazelle only deploys a couple of components of MifosX and backend dependencies we want to have it configurable to deploy all modules e.g. reporting, credit bureau etc at the latest versions

Description

In this project you will update the version of Mifos X included in the Mifos Gazelle release you will include all Mifos X module not just the web-app currently deployed this will require you to be able to check dependencies and also consider infrastructure rationalisation and system constraints across multiple environments.

Skills

cloud native technologies

kubernetes,

helm,

java,

bash scripting

python 

Resources

Mifos Repositories

https://github.com/openMF/mifos-gazelle
https://github.com/openMF/mifos-x

Impact

This project will provide users access to  a much more complete set of MifosX functionality via the simple Mifos Gazelle deployment model

Provisional Mentors

Tom Daly, TBC

Channels to Discuss

Slack: https://mifos.slack.com

Channel: #mifos-gazelle-dev

Overview and Objectives

Update Payment Hub EE Java components to most recent applicable JDK release and update corresponding dependencies such as SpringBoot, camel etc

Description

This project will update the out of date dependencies within Payment Hub EE in particular focussed on but not limited to Java. These updates are essential to the long term sustainability of the DPG. These updates will also be used in Mifos Gazelle which incorporates Payment Hub EE for DPI demos. Whilst this project may sound simple it is not with many repos and interdependencies this project is for someone who is willing to dive deep, use investigative skills and happy that testing is a major part of upgrading of dependencies.

Skills

Java,

SpringBoot ,

Apache Camel,

Gradle build tools 

Resources

Paymenthub repos in https://github.com/openMF

Mifos Repositories

Paymenthub repos in https://github.com/openMF

Impact

This a very important and strategic project which will enable the future development of PaymentHub EE. A longer term approach by building repeatable tests could set the scene for maintainability of PH-EE when dependancies need upgrading in the future.

Provisional Mentors

Tom Daly, David Higgins, TBC

Channels to Discuss

Slack: https://mifos.slack.com

Channel: #payment-hub #mifos-gazelle-dev

Overview and Objectives

This project focuses on extending the existing React based Mifos X web app to make it production ready and a strong candidate to become the primary Mifos X web interface.

It continues the previous development cycle by strengthening Apache (R) Fineract integration and Including the rest of the modules, bringing the React web app features and stability comparable to the current main web (angular) application.

Description

The project involves further development and stabilizing the existing React based Mifos X web application, which is currently under active development and not yet production ready.

The intern will implement the remaining functional modules and complete core workflows such as clients, groups, loans, savings, reports to reach feature similarity with the Angular based Mifos X web app.

The work includes strengthening Apache (R) Fineract API integration using the OpenAPI generated client, adding remaining modules, improving error handling,, and refining the UI using ShadCN and Tailwind.

Skills

React, Typescript, Tailwind CSS, Apache Fineract, ShadCN

Resources

React web app:

React web app report: https://gist.github.com/Craig-Rosario/985588b2576f45686340557d5299c89a

Issues with the current react web app (due to OpenAPI inconsistencies): https://github.com/openMF/mifos-x-web-app-react/blob/dev/ISSUES.md

Mifos Repositories

https://github.com/openMF/mifos-x-web-app-react

Impact

This project will help deliver a modern, production ready web interface for Mifos X, improving usability, and maintainability. By strengthening the React-based web app and connecting it with core Apache (R) Fineract using the OpenAPI client, it can serve as a strong alternative and next step for the future Mifos web interface.

Provisional Mentors

Craig Rosario

Channels to Discuss

Slack: https://mifos.slack.com

Channel: #web-app

Overview and Objectives

Create an extension mechanism that allows to integrate workflow engines

Description

The main goal is to avoid leaking technology details of the specific workflow engine into Apache (R) Fineract. It should be possible to abstract the interactions with the workflow engines with a relatively simple Java interface; in the end the only interaction that needs to happen is to send events (aka input parameters) to the workflow engine to trigger actions/tasks (commands are a good candidate for that). With the new command processing infrastructure it should be fairly easy to map workflow events/actions to Apache (R) Fineract commands with relative small amounts of glue code per specific workflow system. Details still need to be fleshed out. Out of scope: the actual workflow configurations (BPMN, JSON, YAML) don't have to be "write-once-use-anywhere"; abstracting the specific workflow formats into a common file format (like e.g. Serverless Workflow) is very hard and is probably not worth the effort.

Skills

Java, Spring framework, Spring Boot, BPMN, Conductor Workflow Engine (optional)

Resources

• https://cibseven.org/en

• https://github.com/flowable/flowable-engine

• https://github.com/cadence-workflow/cadence

• https://github.com/conductor-oss/conductor

Mifos Repositories

https://github.com/openMF/mifos-workflow

Impact

Big, a lot of "workflows" are currently hard-coded in Apache (R)Fineract. Migrating some/a lot/all of the "if-then-else" parts in the current business logic could help to make Apache (R) Fineract more extensible/customizable and give users a choice what kind of workflow engine they need for their project.

Provisional Mentors

Aleks Vidakovic

Channels to Discuss

Slack: https://mifos.slack.com

Overview and Objectives

This project aims to bridge the gap between complex banking configurations and non-technical users by embedding a Generative AI “Copilot” into the Mifos X Web App. Currently, configuring loan products, tax components, and accounting rules requires deep domain knowledge, creating a steep learning curve for new MFIs. The objective is to build an intelligent assistant that interprets natural language requests (e.g., "Create a weekly loan product with 5% interest") and automatically generates valid, pre-filled configuration forms. Outcome: An in-app widget that dramatically reduces setup time, allowing administrators to configure the system via chat while maintaining a "human-in-the-loop" verification process to ensure safety and accuracy.

Description

The intern will be responsible for end-to-end integration of an LLM-based assistant within the Angular frontend. Key tasks include:

Prompt Engineering: Design and test system prompts that accurately map natural language to Mifos X API payloads and JSON schemas.

RAG Implementation: Implement a lightweight Retrieval-Augmented Generation (RAG) pipeline that grounds the AI in Mifos user documentation to prevent hallucinations (e.g., ensuring interest rates respect platform constraints).

UI Integration: Develop a conversational widget within the Mifos X Angular codebase that accepts user input and renders the "suggested" configuration in the existing form fields for review.

Verification Loop: Build a validation layer that checks the AI-generated JSON against Fineract's validation API before presenting it to the user.

Skills

Angular (TypeScript), Python or Java (for backend orchestration), LLM APIs (OpenAI/Anthropic/Llama), Prompt Engineering, Vector Databases (for RAG), REST APIs.

Resources

Apache Fineract API Documentation: https://fineract.apache.org/

LangChain/Flowise Documentation (for RAG architecture).

Mifos Repositories

Mifos X Web App Repository: https://github.com/openMF/web-app

Impact

Dramatically reduces the "Time-to-Value" for new partners by simplifying the onboarding process. Social: Lowers the barrier to entry for smaller MFIs in developing regions who lack dedicated IT teams, enabling them to digitize operations faster. Technical: Positions Mifos X as a forward-thinking, AI-enabled Digital Public Good.

Provisional Mentors

Yash Sancheti

Channels to Discuss

Slack: https://mifos.slack.com

Channel: #web-app, #wg-ai-for-all

Overview and Objectives

Event driven architecture to improve modular design and decrease module dependencies

Description

Use Spring's application event publisher to as many direct dependencies from large modules as possible. This works best for "one-way" scenarios (read: function calls that have are void/no return values).

Candidate use case is for this project is build out stand-alone modules for savings group support using these event-driven architectural patterns.

Skills

Java, Spring Framework, Spring Boot (optional), clean code (optional)

Resources

n/a

Mifos Repositories