Hybrid LES-Filtered-Density Function (LES-FDF) approach has the key advantage that complex unsteady reacting flows can be simulated without the need for chemical source term modeling. This hybrid approach has been deemed to be computationally intractable due to the size of the computational grids and the need to evolve large ensemble of particles. In addition, the chemical source term for combustion applications can lead to stiff-ODEs that need to be integrated in time for each particle at each time step. To overcome this challenging computational problem, several implementation issues need to be addressed. The aim of the current work is to formulate a high-fidelity LES-FDF algorithm for turbulent combustion. Here a low-Mach number variable density LES solver is coupled to a newly developed particle-based FDF solver. The coupled algorithm uses MPI-based parallelism and is found to provide linear-scaleup upto 128 processors for the cases considered. An efficient face-to-face particle tracking has been implemented to aid in particle evolution across processor boundaries. Due to the unsteady nature of LES, a temporally accurate coupling strategy based on the enthalpy source term feedback is used. Several simple flow configurations are used to test the consistency, accuracy and efficiency of this implementation. Using a 14-step augmented reduced mechanism derived from the GRI-2.11 reaction scheme, an experimental bluff-body stabilized flame is simulated. Direct integration of the chemical source term, as expected, is found to be the most computationally intensive part of the calculation. Some intuitive techniques used to reduce the number of direct integrations were found to accelerate the computation significantly. Detailed comparison with experimental data show excellent agreement and illustrate the future potential of this approach. The effect of sub-filter scalar mixing model on FDF evolution is also discussed.