Complex physical systems frequently undergo conformational transitions between free energy minima separated by a large barrier. The large resultant disparity in timescales for inter-well and intra-well dynamics greatly limits the use of conventional all-atom Molecular Dynamics (MD) simulation to study conformational transitions. To circumvent this difficulty we propose a synergistic combination of coarse molecular dynamics (CMD) and reverse integration to extract, on-the-fly, information (in reaction coordinate space) about the slow dynamics of the system from multiple short microscopic simulations. In this approach, reaction coordinates are extrapolated "backward-in-time" using information from their forward time evolution obtained from the fast relaxation dynamics of the system. We utilize our method to reconstruct the two-dimensional folding free energy surface of a simplified model of src-SH3 domain, a 57 residue protein. The protein is modeled using an off-lattice C_a representation associated with a realistic, minimally frustrated Hamiltonian. Previous experimental and simulation studies of this system indicate that a two-state folding mechanism is operative with a single free energy barrier separating the folded and unfolded state. Additionally, the folding process may be effectively characterized by two empirical, structure-based, reaction coordinates: the number of native contacts formed, Q, and the number of non-native contacts formed, A. Here, we use Q and A to map the underlying effective folding free energy surface of SH3. The location of the top of the folding free energy barrier on the (Q, A) surface determined using our approach is in remarkable agreement with the transition state on the free energy surface obtained from processing simulation data  from  traditional long-time MD. Furthermore, the proposed method identifies the folding transition state region with a significant reduction in the simulation time relative to standard MD.