Iam Grey Mengezi, a member of KAI Lab, where we focus on building practical AI solutions for African challenges. This post shares my experience attending the Deep Learning Indaba 2025 in Rwanda, an event that brought together researchers, engineers, and students from across the continent and beyond. I attended the Indaba to represent KAI Lab and showcase our work on the mwBTFreddy dataset, which is part of our Disaster Management research project aimed at mapping and analysing building damage in Malawi following Cyclone Freddy.

Attending the Deep Learning Indaba 2025 in Rwanda was one of the most transformative experiences of my journey as an early-career researcher in Machine Learning and Artificial Intelligence. I left inspired, informed, and motivated to continue building my skills and contributing to the AI community.

Workshops, Tutorials and Practicals

The workshops and tutorials were well-organized, with clear structures and engaging speakers. One highlight was the Mathematics of Deep Learning tutorial, which revisited essential concepts in Linear Algebra and Calculus and showed their direct applications to deep learning. This session strengthened my foundational understanding and gave me more confidence in applying mathematical tools to ML research.

The Machine Learning Foundations practical was equally impactful. Having access to the notebooks beforehand made the session smooth and easy to follow. We built and trained deep learning models from scratch using NumPy and later explored how to efficiently scale to larger models using frameworks like JAX. This hands-on progression from fundamentals to practical tools bridged theory and application perfectly.

Lightning Talk & Poster Presentation

I had the opportunity to give a short 6-minute lightning talk during the Research in Africa Day dataset spotlight session. I presented the mwBTFreddy dataset, created for disaster management research in Malawi. Later, I showcased the dataset in my poster presentation, which focused on using ML to detect building damage after Cyclone Freddy. Sharing this work with attendees from diverse academic and professional backgrounds was rewarding. I received insightful feedback and recommendations on methods and tools to further strengthen the research. To my delight, the poster was awarded recognition as one of the best posters under the African Datasets category, which was both validating and motivating.

Poster presentation session (left) and poster awards presentation (right)

Networking and Community

The Indaba also provided a unique space to connect with researchers from Google and Google DeepMind. I learned about scholarship opportunities, and received guidance on preparing for competitive interviews. These interactions expanded my perspective on the pathways available for early-career researchers like myself.

I also met fellow Malawians during the event, and we discussed potential collaborations back home. We talked about IndabaX Malawi and they expressed interest in participating and contributing to it. We also explored ways to engage more researchers and students in Malawi to increase participation in such events, which I believe could have an impact on strengthening the local AI community.

Fellow Malawians at Deep Learning Indaba 2025 in Rwanda

Equally inspiring was the sense of community. From keynote talks like Verena Rieser’s discussion on aligning AI with diverse views to the many informal conversations, I was reminded of the importance of context, diversity, and inclusivity in AI development. This spirit of collaboration reflected this year’s theme, Urunana (Hand in hand) - working together to advance AI in Africa. Meeting fellow Malawians and connecting with researchers from across the continent reinforced the importance of building together, hand in hand, for lasting impact.

Booths and Opportunities

The exhibition booths were another highlight. Organizations such as Google, MILA, IBM, InstaDeep, EqualyzAI, and Etihuku showcased their research, internships, and scholarship opportunities. Each booth offered insights into ongoing work in AI and ways for collaboration, training, and career advancement. For me as an early-career researcher, these interactions were invaluable in expanding both knowledge and opportunities.

Cultural Dinner

The fourth day of the event concluded with one of the biggest fun highlights of the Indaba, the Culture Dinner. This was a celebration of Africa’s cultural diversity, where participants wore traditional attire and showcased their country’s music, dances and poetry. It was a reminder of the unity and richness of the African continent, and it added a warm dimension to the otherwise technical and academic environment.

Cultural dinner at Indaba

Reflections and Future Vision

What stood out most was how well-organized the entire event was. From effective communication to logistical support, the Indaba team ensured a seamless experience for participants. The diversity of topics, speakers, and participants made every session enriching. Looking forward, I would love to see even more advanced deep learning topics covered, along with continued emphasis on applied AI solutions for African challenges.

The Deep Learning Indaba 2025 was not just an event, but a key step in my research journey. It strengthened my technical skills, expanded my network, and deepened my appreciation for the role of diversity and context in AI. Most importantly, it left me inspired to keep learning, building, and contributing to the African AI ecosystem. I am grateful to the organizers, speakers, and community for making this experience possible, and I look forward to being part of the Indaba community in the years to come. See you in Nigeria!

The Government AI Readiness Index is an annual report produced by Oxford Insights that seeks to answer the question: how ready is a given government to implement AI in the delivery of public services to their citizens?. Recognized by UNESCO and the G20, it is a trusted tool for policymakers worldwide. The 2024 index examines 40 key indicators across three areas; Government, Technology Sector, and Data & Infrastructure in order to track progress, identify challenges, and uncover opportunities. As governments navigate economic uncertainty and climate change, AI has the potential to enhance decision-making. By providing clear insights, the index helps policymakers harness AI to better serve their citizens. As AI continues to shape the future, initiatives like IndabaX play a crucial role in promoting research and collaboration across Africa.

The IndabaX programme is a growing initiative that brings together African countries to strengthen the machine learning and AI community. Launched in 2018, it has expanded each year, providing a platform for learning, research, and collaboration beyond the annual Deep Learning Indaba conference. In 2024, the programme supported 47 countries, including Algeria, Kenya, Nigeria, South Africa, Egypt, Ghana, and many more, with events that were scheduled across the continent. These gatherings, organized locally, range from small workshops to large conferences with hundreds of attendees, encouraging knowledge sharing and innovation in AI. The initiative continues to grow, aiming to reach every African country in the near future. The table below shows the countries that participated in IndabaX 2024 and their history of hosting the IndabaX event.

Table 1: IndabaX Countries and Event Participation History (2018–2024)

To understand how these insights translate to African AI development, we examine the AI preparedness of countries within the IndabaX community. We use the data provided by Oxford insights to identify which nations are best positioned to implement AI in governance and public services. As IndabaX continues to strengthen AI research and collaboration across Africa, understanding the AI readiness of its member countries provides valuable insights into regional strengths, gaps, and opportunities for growth.

Findings

The map below illustrates the global readiness of countries to adopt AI. Countries least prepared for AI adoption are represented by the color on the far left of the scale, while readiness increases as the color shifts toward the right. The countries with the color on the far right of the scale are the most prepared to implement AI.

Based on the 2024 AI Readiness Index, several IndabaX countries stand out as frontrunners in AI adoption. Egypt emerges as the most AI-ready country within the IndabaX community, ranking 65th globally with a preparedness score of 55.63. It is followed by Mauritius at 69th place with a score of 53.94% and South Africa at 72nd with 52.91%. The following sections break down AI readiness into the three key pillars: government support, technology sector development, and data and infrastructure readiness.

Government Support for AI: Rwanda leads in government support (71.44), followed by Egypt (68.98) and Mauritius (65.31), demonstrating strong AI policy efforts and investment. In contrast, Sudan (13.32), the Central African Republic (12.07), and Guinea-Bissau (14.65) show weak governmental backing.

Technology Sector Development: Egypt (42.13), South Africa (39.15), and Tunisia (41.07) have the strongest technology sectors, promoting AI innovation. Meanwhile, Chad (18.22), DR Congo (15.99), and Angola (15.87) struggle due to underdeveloped tech ecosystems.

Data and Infrastructure Readiness: South Africa (65.28), Mauritius (63.81), and Tunisia (61.35) are best positioned in terms of data availability and infrastructure, essential for AI adoption. However, Burundi (27.84), Chad (28.82), and the Central African Republic (28.77) face significant gaps.

The charts below categorize IndabaX countries into three groups: the top 10 most AI-prepared, mid-level countries, and those that are least prepared. Addressing disparities in government support, technology, and infrastructure will be key to advancing AI readiness across the region.

Chart 1: The top 10 most prepared countries

Chart 2: The mid-level countries in terms of AI preparedness

Chart 3: The least prepared countries

The 2024 Government AI Readiness Index highlights the varying levels of AI preparedness among IndabaX countries. Egypt, Mauritius, and South Africa lead the rankings, demonstrating strong governmental support, a growing technology sector, and developed data infrastructure. Rwanda, Senegal, and Tunisia follow closely, signaling steady progress in AI adoption. However, many nations, particularly those ranking lower on the index, still face challenges in policy development, digital infrastructure, and AI investment. Despite this, the IndabaX community is working towards the goal of Africans not only being observers and receivers of the ongoing advances in AI, but active shapers and owners of these technological advances. With Continued collaboration, investment, and better policies, even the least-prepared countries can improve their AI readiness in the future. Below is an example chart that shows other factors that were used to measure AI readiness of a country.

On the 31st of January 2025, my colleague, Chimwemwe Chamangwana, from KAI Lab and I had the privilege of attending the International Day of Education event at Katoto Secondary School in Mzuzu. Accompanied by a representative from UNIPOD, we joined educators, students, and institutions from across Malawi to engage in discussions about the evolving role of Artificial Intelligence (AI) in education.

The event was graced by two leaders: Hon. Madalitso Kambauwa Wirima, Minister of Basic and Secondary Education, and Hon. Jessie Kabwila, Minister of Higher Education. Both ministers underscored the importance of integrating AI responsibly into Malawi’s educational system, emphasizing the need to preserve human agency in an increasingly automated world. Their speeches resonated with the event’s theme: “Artificial Intelligence and Education: Preserving Human Agency in a World of Automation.”

Showcasing AI Innovations from KAI Lab

As part of our participation, we presented two AI-driven projects that show how technology can enhance education and healthcare accessibility in Malawi.

MUBAS Stand

• Leveraging AI to Model Determinants of Health Care and Education in Malawi: This project focuses on using AI to map schools and healthcare facilities across Malawi, with the aim of identifying underserved areas and improving service delivery. By integrating AI-driven Geographic Information Systems (GIS) and modeling, this project analyzes accessibility challenges, providing data-driven recommendations for infrastructure development. Our work has already mapped over 340 schools and 146 health facilities in Blantyre, offering valuable insights into how AI can optimize resource allocation in education and healthcare.
• The role of AI in self-directed learning and professional development: Developed at KAI Lab in collaboration with the Public Health Institute of Malawi (PHIM), IntelSurv leverages AI to support community health workers in understanding medical terms, case definitions, and data entry requirements. The tool provides real-time AI-powered feedback, improving training, knowledge access, and ultimately the quality of healthcare service delivery.

Engaging with Key Stakeholders

The event began with a tour of various project stands, where the ministers and other dignitaries interacted with presenters. Our team was honored to be among the first to showcase our work, presenting second after the initial display. Representatives from the University of Malawi (UNIMA), Lilongwe University of Agriculture and Natural Resources (LUANAR), Mzuzu University (MZUNI), and various secondary schools were present, fostering a rich exchange of ideas.

Following the presentations, attendees convened in the school hall for a series of activities, including panel discussions and student-led performances, all centered around AI and its transformative impact on education.

School Hall

Key Messages from the Ministers

Dr. Jessie Kabwila encouraged institutions to leverage AI as a means of fostering innovation, ensuring that its application enhances rather than undermines human values. She highlighted the necessity of ethical AI implementation to help reduce socio-economic inequalities in education. Hon. Madalitso Kambauwa Wirima reiterated the government's dedication to embedding AI within national policies to ensure consistency across various sectors. She also announced plans for awareness programs aimed at educating teachers and students on responsible AI use and ethical digital practices.

The Future of AI in Malawian Education

The discussions at this event highlighted the need for AI literacy among students and educators. As Malawi moves towards AI integration in academic institutions, it is crucial to ensure that technology serves as a tool for empowerment rather than replacement. By promoting responsible AI use, we can preserve human agency while leveraging automation to enhance learning and service delivery.

Our participation in this event reaffirmed KAI Lab’s commitment to driving AI innovation for societal impact. We look forward to further collaborations in education, healthcare and other sectors in Malawi.

SSMS fraud, also known as smishing, is a growing global cybercrime where fraudsters exploit mobile users via deceptive text messages to steal sensitive information or money. In African contexts, the challenges are particularly pronounced due to the multilingual environment, frequent code-switching, and a lack of labelled datasets in local languages such as Chichewa. Additionally, the limited availability of linguistic tools and resources further complicates efforts to develop robust detection mechanisms. My early work began with my undergraduate thesis, where I and a colleague collected SMS messages in Chichewa and run machine learning experiments on it. This application was designed to detect and filter fraudulent messages in Chichewa. While the project was promising, gaps in data quality and methodology were identified. The initial dataset, though valuable, needed improvement in terms of size, structure, and diversity, as well as corrections to the methodological approach. Collaboration with Dr. Amelia Taylor allowed for a deep exploration. Sharing a mutual interest in SMS fraud detection, I joined Dr. Taylor to expand the scope of the original project. This collaboration aimed to address critical gaps, including improving the dataset, refining the methodology, and developing a more scalable and accurate detection system tailored for the African context. The expanded research involved several key activities: (1) cleaning and organising the existing dataset from your undergraduate project, (2) collecting new SMS data in Chichewa to create a larger, more representative dataset, (3) refining the methodology to include translation into English to facilitate multilingual model training, and (4) rerunning and improving machine learning experiments to achieve higher accuracy and better generalisation.

This project involved several individuals at the lab, each contributing complementary skills. These included Tamanda Phiri, who was part of the translation team alongside Ben Chapuma and myself. In addition to our roles in translation, Tamanda, Ben and I were actively involved in new data collection efforts which also included Alinafe Lipenga. Dr. Taylor guided us on methodology for machine learning, dataset preparation and data collection.

The inspiration behind the research

Malawi, like many countries, has seen a rise in fraudulent SMS schemes targeting unsuspecting individuals. Several Malawians both in urban and rural areas have fallen victim to this act. A 2023 report from Malawi Communications Regulatory Authority(MACRA) indicated that approximately US$117,000 is stolen every month from Malawians through electronic fraud which usually takes the form of SMS and voice(phone calls). The fraudsters exploit the local language, Chichewa, to reach a wider audience including the rural masses who can read their messages. This Chichewa nature of fraudulent messages makes the traditional fraud detection tools which are often tailored for widely spoken global languages such as English to be less effective. This gap triggered a motivation to explore how machine learning (ML) could be leveraged to address this pressing issue in a local context.

Sample fraudulent SMSs written in Chichewa language

Challenges in working with Chichewa SMS data

One of the first hurdles we faced was the lack of a standardised Chichewa corpus that is suitable for SMSs. Unlike English, where extensive language resources are readily available, Chichewa has limited digital text repositories. There are some Chichewa corpi created for general text: SpokenChichewaCorpus, masakhane-io/lacuna_pos_ner, masakhane-io/masakhane-pos. This scarcity meant we had to get creative with data collection, curating datasets from scratch. We gathered authentic SMS messages, both fraudulent and legitimate, from research participants who had consented and manually annotated them to train our machine learning models.

The role of data augmentation

To enhance our dataset, we employed a manual text augmentation. Chichewa, like other Bantu languages, has rich morphological structures and diverse dialectical variations. By focusing on specific parts of speech such as verbs, pronouns, prepositions, and nouns, we introduced variations in the SMS messages. This process included:

1. Addition of words that were implied by the text but were not explicitly present in the original Chichewa SMS.
2. Synonym replacement through substitution of words with similar meanings to introduce diversity.
3. Replacing borrowed words from English with the equivalent words in vernacular (Chichewa).

Experimenting with machine learning models

Our approach combined natural language processing (NLP) techniques with supervised machine learning algorithms. The preprocessing phase involved tokenising messages, encoding classes and vectorising text using TF-IDF (Term Frequency-Inverse Document Frequency) technique. We trained various models, including logistic regression and random forest, evaluating their performance using accuracy, precision, recall and f1-score metrics.

Insights and outcomes

A subset of 649 messages was used for experimenting with machine learning models. Random Forest model achieved the highest accuracy of 97% while Logistic Regression achieved 96% on Chichewa dataset. Our dataset was also translated by human translators from the research team members and machine translator(Google Translate). From this translated dataset, Logistic Regression achieved highest with 96% accuracy while Random Forest came second with 95%. The performance of both models slightly decreased on the machine translated dataset. Beyond its technical success, the project underscored the importance of developing localized solutions. By tailoring our approach to the nuances of Chichewa, we demonstrated that machine learning can be a powerful tool for addressing language-specific challenges.

A table of machine learning experimental results

New large dataset

The research culminated in a model capable of detecting fraudulent SMS messages with a high degree of accuracy. A total of 15,229 raw SMS messages ,both fraudulent and non-fraudulent, were collected during the data collection exercise. This dataset is currently being prepared for analysis.

Reflections on the experience

This research was more than an academic pursuit; it was a profound learning experience that extended far beyond technical model building. It highlighted the critical role of interdisciplinary collaboration, particularly through engaging with community stakeholders during data collection. Their firsthand experiences and insights provided invaluable context, deepening our understanding of the nuances of SMS fraud in Chichewa. These interactions helped us identify the tactics used by fraudsters and the cultural or linguistic cues that often made these scams convincing, ultimately shaping our approach to designing feasible research methodology.

Additionally, this project reinforced the importance of ethical considerations when working with sensitive data. We prioritised data privacy and security, ensuring that all collected information was handled responsibly to protect participants' identities. These measures aligned with our commitment to maintaining trust and upholding ethical research standards.

Finally, this research emphasised the need for context-aware innovation. By incorporating insights from the communities directly affected by SMS fraud, we shaped an approach that was both practical and grounded in real-world experiences. This ensured that our model not only performed well technically but also addressed the problem in a way that could resonate with and benefit the people it was intended to serve if integrated into a broader solution.

Moving forward

The fight against SMS fraud is far from over. While our model provides a robust starting point, there’s ample room for improvement. Future work could include experimenting with large SMS datasets and other SMS features such as phone numbers, time etc on top of SMS body which was only the focus feature in our research. Additionally, the development of a standardized Chichewa corpus could significantly benefit not only fraud detection but also other NLP applications.

Conclusion

Our journey in detecting SMS fraud in Chichewa illustrates the transformative potential of machine learning when applied thoughtfully to local contexts. By addressing language-specific challenges, we can create solutions that are not only innovative but also impactful. This project stands as a testament to the power of technology to bridge gaps and solve pressing societal problems, one message at a time.

T he devastating impact of Tropical Cyclone Freddy on Southern Africa in 2023 (from February - March) highlighted the urgent need for innovative solutions in disaster management. In Malawi alone, over 500 lives were lost, thousands were displaced and communities struggled to rebuild amidst widespread flooding. While established disaster response methods are crucial, they can be slow or limited in their ability to respond to rapidly evolving crises. During the flooding, social media platforms like WhatsApp, Twitter, and Facebook became vital channels for real-time data sharing, providing an untapped opportunity to bridge these gaps. By leveraging social media data, we can enhance the speed and effectiveness of flood mapping and disaster response efforts, ensuring a more responsive and informed approach to future disasters.

Recognizing this potential, we participated in the 2024 Geography Awareness Week Hackathon under the theme "Mapping Minds, Shaping the World." The event brought together developers, researchers, and mappers to explore how social media feeds could be harnessed to create structured systems for flood mapping. With the devastation of Cyclone Freddy serving as our problem statement and motivation, our mission was clear: to collaboratively design a solution that integrates data from social media into disaster management frameworks, leveraging tools like OpenStreetMap and AI technologies.

What followed was an inspiring day of brainstorming, coding, and problem-solving. We divided ourselves into three teams based on individual areas of expertise, with everyone free to choose which group they wanted to participate in. Each team had a designated leader to guide the process and ensure smooth collaboration. Below is an outline of the tasks each team was assigned to work on during the hackathon:

Team 1: User Interface and Project Development

The first team was tasked with building the core structure of the project. Their focus was on developing a user-friendly interface that would allow for account creation, with affiliation as a required field. They were also responsible for structuring the project to transition through three key stages:

1. Harvesting – Collecting social media posts related to flooding events.
2. Filtering – Ensuring that the collected posts met eligibility criteria, such as including images.
3. Mapping – Organizing the filtered posts and creating a visual map of the flooding data for analysis.

Team 2: Data Filtering and AI Model Development

This team’s focus was on filtering the social media data to ensure it met specific criteria. They were tasked with filtering out posts that did not meet the required standards, such as posts without images. In addition, they were asked to identify duplicate posts to avoid redundancy. Their role also included creating a simple page where users could manually mark posts as relevant or not, aiding in the manual filtering process. Additionally, this team worked on laying the groundwork for developing an AI model to automatically filter out off-topic posts, with the goal of continually training this model as new data was collected.

Team 3: Mapping and Location Automation

The third team was responsible for mapping the filtered social media posts by identifying their actual locations.They explored the use of geographic tools that could automate the process of linking images to specific locations, leveraging an already available dataset of social media posts from Nkhotakota dwangwa floods. Their task also included using Natural Language Processing (NLP) to aid in extracting geographic details from the posts, and cross-referencing them with platforms like Google Maps or OpenStreetMap to determine the exact coordinates of the flooding events.

Turning Vision into Reality: Our Achievements in One Hour

After dividing ourselves into teams and setting clear objectives, we gave ourselves one hour to achieve the tasks. Despite the limited time, we made impressive progress on our goals. Here's a look at what each team was able to accomplish during the hackathon:

Team 1: User Interface Development

The first team worked diligently to lay the foundation for the project’s user interface. By the end of the hour, they had successfully developed a landing page, sign-in page, login page, and project page. Although the time constraint limited the scope of their work, the team managed to create the core structure needed for user interaction and project navigation. This solid starting point will enable further development and refinement in the future.

Team 2: Data Filtering and AI Model Development

Team 2 focused on defining relevance and filtering out irrelevant data. They established clear criteria for what constitutes relevant content, such as the inclusion of images, location-specific details (e.g., landmarks), and time stamps. Additionally, they developed key terminology for identifying flood impacts, actions taken, community engagement, and flood-prone areas in Malawi.

To put this work into practice, the team generated 150 records of non disaster context using ChatGPT, and merged it with an existing dataset containing 167 disaster records. They then trained a random forest model for text(record) classification, though the model’s performance was less than optimal due to a few challenges:

• Less data
• Poor quality of the augmented data
• Selection of the most appropriate model for text classification

Despite these issues, the team gained valuable insights into improving the model’s performance and had a clear path forward for fine-tuning their approach.

Team 3: Mapping and Location Identification

Team 3 focused on mapping the images and text from the Nkhotakota and Dwangwa flood datasets. They had a dataset of 33 images and using satellite imagery, they manually identified 16 images from the 33 suitable to be used for mapping, they manually pinpointed the locations of these 16 images. However, the team faced a challenge in obtaining longitude and latitude coordinates, as the required API cost $300, which was beyond their budget.

They also explored alternatives like Esri which can be used for extracting geographic details from the text accompanying the posts. By analyzing keywords such as "Dwangwa" and "madzi osefukira ku Dwangwa," they identified potential flood locations and used Google to verify their existence. While they couldn’t complete the entire process within the hour, the team also made progress on developing an algorithm that could automate this mapping process in the future.

Team 3 discovered two promising approaches for future flood mapping:

1. Using images to identify flood locations.
2. Using text and keywords from social media posts to pinpoint areas affected by flooding.

In conclusion, the hackathon was a remarkable exercise in collaboration, problem-solving, and innovation. While we couldn’t achieve what we hoped to within the short timeframe, the progress made was significant. Our work during the event laid a solid foundation for future development of a comprehensive flood mapping system using social media data. With these insights, we are one step closer to developing a more efficient and responsive disaster management framework, capable of leveraging real-time data for better flood mapping, quicker responses, and ultimately, a more resilient future.

I had a great summer this year; from being with family to seeing friends I hadn’t seen in a while. One of the highlights of my summer was working as a research intern at KAI Lab at the Malawi University of Business and Applied Sciences (MUBAS). I had seen LinkedIn posts of the wonderful work the Lab is doing with AI to solve problems in the Malawian context. As a firm believer that AI bias can only be solved by building AI solutions in the local context, KAI Lab is one of the places doing that. An example of a project that really caught my eye was their SMS Fraud Detection using Machine Learning tool. With SMS Fraud prevalent in Malawi, this was solving a real problem. I wanted to be part of the AI work that is happening in Malawi and what better place to do than KAI Lab.

My research revolves around using big data for development in the context of the developing world and KAI Lab provided an environment for that. My first meeting with the team at KAI Lab did not feel like a first meeting at all, everyone was relatable and within minutes we were all throwing jokes and laughing while bouncing ideas off each other. Everyone was open to bringing in and critiquing ideas in our brainstorming meetings. The organizational culture of the internship environment was characterized by a strong emphasis on family and friendship, contributing to a relaxing working environment. One gesture of family and friendship that left an impression on me was when one of the team members brought me lunch on my second day. This relaxed environment also made it possible to try and learn new things without being afraid of making mistakes. For example, in one of the projects, none of us had experience with QGIS but we decided to take on the challenge and learn. I think this is important for team growth especially in Tech where things are always changing.

I also learned a lot from Dr. Taylor and one of the things I learned was being organized when doing research. Normally, I am a spontaneous person who “throws himself” into something without having a proper detailed plan. Working under Dr Taylor, I learned planning and thorough documentation. This is the “boring” part of the work but I learnt that it saves you a lot of time as research progresses. By ensuring that steps are documented before implementation, the team is provided with greater clarity and direction, streamlining the workflow and fostering a more cohesive and effective team dynamic. I also learned to work in a team setting doing research.My goal is to do a Phd and this experience is valuable considering that I do not have a lot of research experience. As a lead of one of the projects, which involved creating a dataset for the effects of cyclone Freddy in Malawi to be used for object detection and building damage classification, it was a learning experience leading a team in research and figuring things together considering the data scarcity problem in developing countries. We made a lot of mistakes as a team but I learned that “a team that makes mistakes together, stays together” as we would laugh at ourselves and collectively correct our mistakes as a team.

Working at KAI Lab deepened my research skills, plugged me into a family of researchers, and gave a research mentor in the process. I am excited about the work happening at the Lab and I look forward to working with the lab again. And thanks to Dr. Taylor for this initiative which is relevant as AI becomes prevalent.

My journey throughout these years as an intern at Kuyesera AI Lab (KAI) at the Malawi University of Business and Applied Sciences has been a transformative experience for me. Through the support and resources provided by Dr. Amelia Taylor, I've not only enhanced my skills but also achieved significant milestones, including winning a prestigious competition and securing an opportunity to go to the USA for an attachment. Here's how this incredible journey unfolded. I will share more on how I joined the lab, my first projects, the projects I have worked on with the team, the skills I have learned, the usefulness of this experience and some of my personal highlights. I really hope my journey will inspire and inform those considering a similar path.

How did you join?

I discovered the opportunity through social media. One of the friends who works within MUBAS shared with me a post that they are looking for data annotators. I decided to apply not only because I needed money at that point but also because I really wanted to know what it is like when they say annotation, what do they do? and what can I learn from being an annotator? After submitting my application, I was called for an interview where I was selected to be one of the members in the team.

What was the first project you worked on?

The first my role was to identify and tag names of entities, such as names of towns, names of people, names of countries etc. as words from a given document and provide explanations, and meanings. This was a document full of Chichewa sentences.

This experience has been incredibly useful to me in several ways. Professionally, it has provided me with practical, hands-on experience in IT, allowing me to apply theoretical knowledge to real-world scenarios

What skill did you learn?

OK, so for the first time I did not understand why we are doing all those annotations and I had to ask how they are going to make the annotations useful. When I was told that Dr. Amelia Taylor and colleagues are on a journey of developing an automatic tool that will be able to understand different languages and that they are going to feed all this data for the tool to work. I began to get more interested in machine learning and getting its skills.

I have also improved both my written and verbal skills by working on this task and subsequent tasks. I developed problem solving abilities, and also learned how to be creative in whatever task given.

I am also grateful and thankful for being able to have a skill of collecting data, entering data, cleaning the data and making sure that the data is presentable.

Other skills I have learned are programming languages e.g. Python.

How useful has this experience been to you?

This experience has been incredibly useful to me in several ways. Professionally, it has provided me with practical, hands-on experience in IT, allowing me to apply theoretical knowledge to real-world scenarios. Academically, the organization’s support played a crucial role in my success at the RICE 360 and MUBAS competition where I applied the skills and insights I had developed. Winning this competition was a remarkable milestone, and it opened the door to an exciting opportunity: an academic attachment in the USA.

What are your highlights?

Some of the most memorable highlights of my journey with KAI Lab as an intern include:

Successfully completing other projects together with the team, such as the development of a catalog of news and journal articles, testing Large Language Model tools. Contributing to these projects was a significant achievement, showcasing my ability to do impactful work.

Winning Rice 360 annual competition: The skills and experiences I gained through the Kuyesera AI Lab were instrumental in securing this victory.

Securing an Attachment in the USA: This prestigious opportunity will allow me to do my attachments and professional development on an international level, thanks to the foundation built by my time at KAI LAB.

Conclusion

The journey with KAI LAB has been nothing short of transformative. The skills, knowledge, and support I received not only enhanced my academic performance but also led to incredible opportunities like winning a major competition and securing an academic attachment in the USA. I am immensely grateful for the experiences and guidance provided by Dr. Amelia Taylor and colleagues at the lab, and I look forward to leveraging these opportunities to further my academic and professional aspirations. If you're considering joining KAI LAB as an intern, I wholeheartedly encourage you to take the step—it could be the start of an equally rewarding journey for you.

What is Metadata?

Metadata refers to data that provides information about other data. In publishing, metadata refers to various details about an article or a book, such as the title, author(s), DOI, publication year, abstract, keywords, and sources. It serves as a key component, complementing the content itself. Properly curated metadata is essential for facilitating data discovery, comprehension, and organization of collected journal articles. Additionally, it simplifies communication with authors or affiliated institutions, thereby enhancing the research process for scholars.

What is MetaFetch?

In many cases, researchers collect files containing published articles but may lack essential metadata fields such as all the authors, publication years, abstracts, or keywords. To fill in these gaps typically, one might resort to manually searching for each article on various publishing platforms to gather the missing information. However, this process becomes tedious and error-prone, especially with large datasets.

An alternative method involves leveraging public APIs offered by many publishing companies to programmatically fetch the missing metadata. This approach automates and streamlines the process of retrieving metadata for multiple articles. Inspired by this streamlined approach, Metafetch specifically automates the retrieval of metadata for journal articles from multiple sources.

Currently the tool uses the APIs to retrieve the metadata from Elservier’s Scopus, PubMed and CrossRef.

How does MetaFetch work?

Given a file of journal articles which may contain some gaps in metadata, the tool uses APIs from journal sources to retrieve the missing data. The MetaFetch interface is easy to navigate and has instructions the user can follow. Here is an overview of the steps you can take to use the tool.

1. Prepare the input file by using the same field names as specified in the provided sample files. The input file is required to have the article title or doi to be used as the search subject, one of these must be present in the file.
   
   
   
   
   
   
   
   
   
2. Upload the file thus prepared using the interface. You can select a specific API representing the source. If you are not sure of which one to use, select the ‘ALL’ option.
3. After the file selection, a summary of the relevant gaps in the input will be presented.
4. Click the ‘Fetch Metadata’ to get the data.
5. Download returned data files by clicking the buttons;
   1. ‘Merged file’ contains input file data merged with the returned data.
   2. ‘Raw file’ contains only the returned data by the API’s.

Key Features

1. User-Friendly Interface: The UI is easy to navigate through and gets you started instantly.
2. Efficiency: Can fetch and process a batch of 100 articles in under 5 minutes.
3. Accuracy: Designed with accuracy in mind. Only 100% matches are returned.
4. Comprehensiveness: The tool uses three of the richest publishing databases namely Elservier’s Scopus, Pubmed and CrossRef to ensure wide coverage.

Where do we go from here?

While acknowledging the capabilities of the tool, it's important to consider more questions it could address. For instance, can we improve its comprehensiveness by integrating additional APIs from various sources? Furthermore, is it possible to extend its functionality beyond mere metadata retrieval to include information extraction capabilities?

These inquiries serve as the foundation for advancing the tool. By incorporating more APIs, we can enrich the tool's capabilities with a wider range of data sources. Additionally, by including information extraction functionalities, we can empower users to get more insights from the content itself.

Have you ever wondered what strategies could effectively address the challenges of limited internet accessibility and reliability in Malawi?" The Internet Society (ISOC), in collaboration with the Malawi Internet Service Providers' Association (MISPA), the Internet Corporation for Assigned Names and Numbers (ICANN), and Packet Clearing House (PCH) jointly organized a workshop in Mangochi, at Nkopola Sunbird Hotel from the 3rd of April to the 5th of April 2024 held with network operators and service providers responsible for data transmission in Malawi. The main objective of the workshop was to raise awareness on key and important technical and non-technical aspects that ensure the Internet in Malawi is robust, reliable and resilient, and provide a broad understanding of internet infrastructure in Malawi to teach (advise) companies how to improve their services and collaborate with each other towards building a modern Malawi and a modern world (online world, where people communicate, create, and share information and content through digital channels and devices) in the near future.

The agenda of the three days workshop

I attended this workshop as a representative of the KAIL Lab where I worked, among other things, on research that looked into the “Internet infrastructure and Usage in African Deep Learning Indaba participating countries”. Therefore, at this event I wanted to gather more insights from network operators and service providers in Malawi about challenges facing Malawi’s internet infrastructure. This was a well attended workshop with participants from several countries: Malawi, America, Benin, Rwanda and South Africa. There were 40 attendees representing companies & institutions such as PCH, ISOC, ICANN, Airtel Malawi, TNM, MTL, ctn, INK, Global internet Malawi Limited, MACRA, MUBAS, KUHES, UNIMA who are operating in the digital or educational space. There were three main speakers namely; Nishal Goburdhan from Packet Clearing House (PCH) who was also the guest of honor, Ghislain Keramugaba from the Internet Society (ISOC) who was also the workshop coordinator and Yazid Akanho from Internet Corporation for Assigned Names and Numbers(ICANN). The chairperson of MISPA, Dr. Paulos Nyirenda who chairs the management team for the Malawi Internet Exchange (MIX) opened the workshop with welcoming remarks and declared the workshop officially open.

On the first day, Mr. Nishal Goburdhan started by laying out some of the problems that the Malawi Internet is facing (Internet being expensive in Malawi, Lack of a country’s Internet connection, many complaints about network issues and complaints about internet abuse online) and talked about the importance of internet exchange points (IXP) and how to use them to better in the networks in Malawi. His talk helped me understand what an Internet exchange is, the fundamentals of IXP operations, the value of peering, peering coordinators, traffic analysis & peering policies, and how to build strong technical communities.

On the second day, Mr. Yazid Akanho gave an overview of the DNS and BGP services, the number of routers that we have in the world, how they work, how data is transmitted via these routers, and how the recursive server (also commonly known as the DNS resolver, has the important responsibility of seeking requested data and responding to users’ DNS queries) communicates to the routers etc. His presentation helped me to realize the importance of validating and monitoring systems in order to provide excellent services to customers.

On the third day we had an intense discussion where companies were asking practical questions. The discussions were led by Mr. Ghislain Keramugaba. Participants shared their experiences in the field of networking, expressed challenges they have faced and how they solved them, and the tools their companies use in order to provide better services to their customers.

My key takeaways from this workshop includes the idea that we need to take baby steps in order to solve bigger problems, always monitor and validate systems, learn new things by implementing them, always think of how to make systems better and prioritize attending as many workshops and conferences as this is where many opportunities are presented. The workshop ended with us agreeing to collaborate with each other by contributing money to build more exchange points in Malawi, write to Macra requesting for some rules and costs related to internet service providence to be revised and hold more meetings in order to improve the internet infrastructure in Malawi.

The response of Malawi to COVID-19 highlighted the significance of surveillance data in shaping effective disease control strategies. The current system involves gathering client data manually by district health workers using paper-based COVID-19 Case Based Surveillance (CBSR) forms. This information is consolidated in an Excel line list database submitted to the Public Health Institute of Malawi (PHIM). The shared data is further consolidated with other district data for public dissemination and disease response planning.

In 2020, before the pandemic escalated, the District Health Management Team (DHMT) facilitated the training of health workers in COVID-19 case management and response. The team included Environmental Health Officers, Laboratory Officers, Clinicians, nurses, and Health Surveillance Assistants (HSA) who were tasked with managing COVID-19 cases and executing surveillance tasks. Unfortunately, at that time, there was very little information available, and the training was not comprehensive, lacking data management guidance or in- depth surveillance training. Nevertheless, the rapid response team grew and played a vital role in patient care, collecting client data and entering it into the Case-based Surveillance forms. Despite these efforts, the data collection system faced numerous challenges, such as understaffing and being overwhelmed by the growing number of COVID-19 cases and clients requiring testing and review.

To support the surveillance efforts, interns were incorporated into the Environmental Health team. They were provided with on-the-job data collection and management training, despite the knowledge limitations in using the data collection tools. The teams worked tremendously hard to screen and manage COVID-19 cases in prisons, schools, hospitals, and other areas, but the nature of the disease and the information given out constantly evolved. The DHMT made efforts to train health workers in COVID-19 case management with updated guidelines from WHO and the Ministry of Health, but there was no formal training on using the Case- Based Surveillance form and key data management principles.

This project is an excellent solution to the challenges districts have faced with data for many years, and its impact will be tremendous in any future pandemic.

The Lilongwe DHMT received concerns from PHIM about the quality of data submitted and incidents of data mismatches in national and district data disseminated to the public that frequently occurred during the pandemic. It was apparent that there were flaws in the data collection, management, and reporting process. DHMT worked with the surveillance teams to address the issues raised. Still, the manual data entry systems, lack of training, understaffing, and workload remained vital matters that needed to be addressed to combat the problems.

Throughout the pandemic, the challenges highlighted as potential causes of data discrepancies in the district were only assumed, and there was never any data to support the assumptions and to warrant action. It wasn't until 2022, when the PEACH project led by Dr Amelia Taylor started exploring the national COVID-19 list data, that the data evidence gap was filled. The data discrepancies were numerous, including missing data, incomplete data, and incorrect terminologies. We collaborated and began to investigate the data collection and management systems and reasons for data discrepancies in Lilongwe and Blantyre districts through a qualitative research project. Our study findings highlighted the need for training health workers in the use of the case-based surveillance form and developing a feedback mechanism to address data discrepancies in the line list.

The qualitative research findings led to the development of the Intelligent Surveillance Project (IntelSurv), which leverages large language models to develop a Surveillance training application and create an intelligent surveillance feedback system. This project is an excellent solution to the challenges districts have faced with data for many years, and its impact will be tremendous in any future pandemic. Health workers will have a quick and accessible tool that will enable them to collect data correctly, and the feedback system will reduce the workload that the manual system poses on health workers.

As a member of the District Health Management team for Lilongwe and lead for Covid-19 case management and Cholera, I am thrilled about this project and the transformative work it is offering to districts. We have had a long journey of numerous data challenges, but the Intelligent Surveillance Project has shown us that there is hope in changing this by leveraging the existence of Artificial Intelligence.