Unemployment constitutes one of the major problems in developing countries, with factors such as unavailable skills and the proliferation of unskilled workers being cited as main causes. The existing literature has shown that using social networks analysis techniques, employment options for students and graduates can be accurately generated through recommender systems (RS) to optimize the chances of employment. Therefore, this study proposes a fine-tuned Bidirectional Encoder Representations from Transformers (BERT) model named Contextual BERT (CoBERT) for item-centered content-based (CB) filtering RS. The model uses socio-political background, graduate-employer relationship and graduate academic competencies of information technology (IT) students as contextual factors to recommend employability profiles of IT students in unstable developing countries (UDCs) such as the Democratic Republic of Congo (DRC). The performance of the proposed model is verified using four metrics. To evaluate the online performance of the model, the metrics of Diversity and Novelty are used. The metrics of normalized discounted cumulative gain (nDCG) and mean average precision (MAP) are applied to evaluate the model offline. The proposed model shows a high nDCG value at Top-N=4, equaling to 0.99. Validation statistics of 0.5 MAP score and significant t-statistics test values for Diversity and Novelty indicate that the proposed model can be generalized in the DRC and other similar UDCs. The RS model proposed in the current study contributes to the literature on employability models by advancing beyond traditional approaches such as fuzzy logic (FL), and also by taking graduate skills into account. The proposed model is a novel employability recommendation technique that provides highly precise and contextual recommendations. The output from this study contributes a theoretical foundation for implementing RSs in education and the industry.