Introducing Rafiqi 2.0: How artificial intelligence can be key to refugee integration?

In 2016, I started Rafiqi as a side project aiming to connect refugees to mentors and opportunities, with the ultimate goal of facilitating their integration. In the middle of the heated debate on refugees in Europe, my goal was to act, even modestly, to bridge the economic and cultural gap between newcomers and host communities. Despite my middle eastern migrant background, starting this project made me realize how modest my understanding of the refugee space was. It took me more than a year of exploration, meeting with newcomers in Amsterdam (where I used to live) and listening to their struggles and aspirations, meeting with refugee aid organizations and individuals who gave up their jobs to start refugee-focused initiatives, in order to start to understand the diversity and complexity of the refugee space. In the process, some work was done: almost 30 newcomers were matched to mentors and opportunities, mostly manually using my personal connections (to mentees, mentors and opportunity providers). Not all the matchings actually led to an impact but both success stories and less successful matchings were an opportunity to learn. Over a year of encounters with refugee and refugee-focused initiatives made me more optimistic about refugee integration prospects, especially in countries like the Netherlands and Germany. It also made me aware of some facts that can hinder the realization of these prospects, including:

1. Lack of a single platform where refugees can access and browse the wide catalog of opportunities available to them in their host countries, and where any NGO or company can access and filter refugee talent.
2. A refugee’s exposure to opportunities depends of where they live and of the size of their social network. This results in some refugees being exposed to a wide spectrum of opportunities while others are presented with none. This also results in situations where refugees follow a training or do a job which is not in line with their profiles for the simple reason that any opportunity is often preferred to none.
3. Refugees are not exposed enough to a wide range of non-refugee labeled opportunities. These include free world-class trainings provided through e-learning websites or online universities, online freelancing jobs, and trainings provided to women, young people or people with migrant background by local municipalities.
4. Matching a refugee to a mentor with no clear end goal (opportunity to seize) in mind does not often lead to a concrete outcome.

I always believed that marrying tech and social work can bring a huge benefit to humanity. Sadly, working in big tech companies and working in parallel on social projects made me realize that people in NGOs rarely understand the language of tech and its trends, while techies rarely get interested in diving deep into the social impact space.

I do strongly believe that tech (machine learning in particular) can be of great benefit in the refugee context due to the large diversity of refugee profiles and of opportunities available to refugees. A refugee who speaks English and has a strong coding background does not need the same kind of help as a refugee with poor english and no university education. Similarly, an organization providing trainings on entrepreneurship should not be presented with applicants with no interest in becoming entrepreneurs, and an organization providing advanced coding courses should not be presented with people with 0 coding background. Manually finding the right matches reveals to be a challenging and lengthy process, especially when the number and diversity of applicants and opportunities exponentially increases.

So, how data & machine learning can be of help?

Last November, I changed cities and jobs. In the middle of all these exciting and tiring changes, Rafiqi was still in the back of my mind. I used my resettling time to explore the above question. Below are my findings and the subsequent steps that I took:

1. Clustering opportunities available to refugees is an essential starting point. Moving from raw information about different opportunities into a structured data format (also referred to as data cleaning and preparation) is as important, if not more, as doing a successful matching. I sat down and put on an excel sheet all the opportunities that I know of and divided these into categories (university degree, job, technical/cultural training, advice…), delivery modes (online, onsite, hybrid), geographic footprint (if applicable), themes (software development, digital education, economics, entrepreneurship education…), difficulty level, language requirements (both english and local language), and education requirements. I did come back to the excel sheet many times and eventually ended up with over 130 opportunities in 9 countries in addition to global ones (available online for everyone)
2. Next Step was to think about basic data inputs from refugees that can inform the matching process. I recalled the learnings from my 1+ year of refugee space exploration and from previous matching and came up with a list of inputs. These include: current country of residence, education level, work & education background, language level, digital ease, job readiness, and interest in entrepreneurship (if the person in question was not already an entrepreneur).
3. After specifying data entries related to refugees and opportunities, I started exploring the matching process and the logic behind. The most intuitive and common-sense approach that came to my mind was a decision tree taking refugee data inputs as tree nodes and different opportunities as tree leaves. In simpler terms, different elements of refugee data are used for better refining and directing opportunities selection till eventually reaching one or more opportunities considered as the most suitable for a given refugee profile at a given time. An example of the logic can be seen in Fig 1 (this is a small chunk of the tree). The entire tree has 30 nodes. The matching algorithm was coded in R. Although the first version consists of a hard-coded tree, the goal is to gather feedback on the matching outcome in order to eventually make the matching algorithm dynamic and self-adaptive.
4. As an algorithm is not necessarily user-friendly, a UI needs to be built on top of it. As I lack talent when it comes to designing and developing UIs, I decided to use Shiny to build a simple Web application to showcase the idea. Rafiqi matching tool can be found here and can also be accessed via the main website. It allows a customized, real time matching of a refugee to the opportunities considered as the most suitable for his/her profile. It also allows an instant evaluation of the service to enhance the matching algorithm in the future and make it self-adaptive.

Fig 1: A sketch of how Rafiqi matching algorithm works

Where to take Rafiqi from here? And how can you help?

Automating Rafiqi is definitely a step forward. However, a lot of work still needs to be done. Below is the type of help needed:

1. Identifying and clustering new opportunities: If you or an organization that you are representing are interested in providing any kind of opportunities for refugees (Education, training, employment, career coaching..), please fill this form . If you know of an organization/individual and think you have enough information, also please fill the form.
2. Doing the marketing and PR work: A PR and marketing rockstar is urgently needed! Marketing Rafiqi mainly consists of making Rafiqi the platform of choice for young refugees looking for their next opportunity to thrive, and for any organization or individual interested in nurturing or accessing refugee talent. Marketing and PR work also consists of managing Rafiqi social media channels (mainly Facebook), and investigating possible business models or funds. If this speaks to you, please feel free to reach out here.
3. Enhancing the current website and matching interface: A volunteer web developer/designer with Javascript experience is needed to help make the website and the matching interface more attractive and reflective of the service use/impact. If this speaks to you, please feel free to reach out here.
4. Building the Rafiqi mobile app: Mobile developers for Android and IOS are needed to build Rafiqi mobile app. If you feel like developing the app as a side project or as part of your studies (this could be an end of year/semester project for a group of CS students), please feel free to reach out here.

Rafiqi was born as a vague idea out of the urgency to do something about a crisis that encompasses us all. As time passed by and my understanding of the refugee space matured, I decided to combine my hard skills with my human responsibility and built a tool that holds the promise of helping newcomers become a step closer to their dreams. Rafiqi means my companion in Arabic. In the era of Artificial Intelligence, it aims to provide an automated companionship and guidance to those seeking a better tomorrow. I hope to find, among the readers of this blog, passionate technologists and humanists interested in becoming my companions in this journey.