I propose a novel blended finance framework for micro, small, and medium enterprise (MSME) lending in fragile African economies, where the conventional static first-loss capital structure and fixed currency hedging ratio are replaced by a dynamic, state-contingent mechanism. The core innovation is a stochastic optimal control policy derived from a Hamilton-Jacobi-Bellman equation, which jointly optimizes the size of the concessional first-loss tranche and the hedging ratio of a local currency derivative instrument. The system responds in real time to four key state variables: the outstanding loan portfolio, the spot exchange rate, the stochastic volatility of the exchange rate, and the domestic credit spread. These variables follow coupled stochastic differential equations, featuring a Heston-type volatility process and an Ornstein-Uhlenbeck spread process calibrated to historical data from three fragile African economies. The aim is to reduce the expected discounted default losses, with the condition that the probability of loss for the senior tranche stays below a predefined threshold. A deep neural network, trained on ten million simulated Monte Carlo trajectories, approximates the optimal control policy and is subsequently deployed to generate daily recommendations for adjusting the first-loss reserve and executing synthetic local currency swaps. This methodology radically transforms the risk-absorption and hedging mechanisms within a dedicated MSME debt fund, substituting predetermined ratios with a unified, condition-dependent decision protocol. The technical assistance facility and loan origination platform persist, yet their functioning is now guided by the dynamic hedging state, which establishes a stabilizing feedback loop. Our contribution is a mathematically rigorous, data-driven approach to blending concessional and private capital in high-volatility environments, thereby reducing the systemic risk of currency crises and promoting more sustainable MSME lending in the world’s most fragile economies.