<p dir="ltr">Quantum computing is advancing rapidly, with variational quantum algorithms (VQAs) showing great promise for demonstrating quantum advantage on near-term devices. A critical component of VQAs is the ansatz, a parameterized quantum circuit that is iteratively optimized. However, compiling ansatz circuits for specific quantum hardware is challenging due to topological constraints, gate errors, and decoherence. This thesis presents a series of techniques to efficiently generate and optimize quantum circuits, with a focus on VQAs. We first introduce AccQOC, a framework combining static pre-compilation with accelerated dynamic compilation to transform quantum gates to hardware pulses using quantum optimal control (QOC). AccQOC generates pulses for frequently used gate sequences in advance and stores them in a lookup table. For new gate sequences, it utilizes a Minimum Spanning Tree based approach to find the optimal compilation order that maximizes the similarity between consecutive sequences, thereby accelerating the compilation process. By leveraging pre-computed pulses and employing a similarity-based approach, AccQOC achieves a 9.88×speedup in compilation time compared to standard QOC methods while maintaining a 2.43×latency reduction over gate-based compilation. Building on AccQOC, we propose EPOC, an extended framework integrating circuit partitioning, ZX-calculus optimization, and synthesis methods. EPOC operates at a finer granularity compared to previous coarse-grained approaches, decomposing circuits into smaller sub-circuits based on the number of qubits and circuit depth. It then applies synthesis techniques to identify equivalent representations with reduced gate count. The optimized sub-circuits are then grouped into larger unitary matrices, which are used as inputs for QOC. This approach enables increased parallelism and reduced latency in the resulting quantum pulses. Compared to the state-of-the-art pulse optimization framework, EPOC achieves a 31.74% reduction in circuit latency and a 76.80% reduction compared to gate-based methods. To construct hardware-efficient ansatz for VQAs, we introduce two novel approaches. TopGen is a topology-aware bottom-up approach that generates sub-circuits according to the connectivity constraints of the target quantum device. It starts by generating a library of subcircuits that are compatible with the device topology and evaluates them based on metrics 17 like expressibility and entangling capability. The sub-circuits with the best properties are then selected and progressively combined using different techniques. TopGen also employs dynamic circuit growth, where small sub-circuits are appended to the ansatz during training, and gate pruning, which removes gates with small parameters. Evaluated on a range of VQA tasks, TopGen achieves an average reduction of 50% in circuit depth after compilation compared to previous methods. NAPA takes a more direct approach by utilizing devicenative parametric pulses as the fundamental building blocks for constructing the ansatz. It uses cross-resonance pulses for entangling qubits and DRAG pulses for single-qubit rotations. The ansatz is constructed in a hardware-efficient manner. By using the better flexibility and expressivity of parametric pulses, NAPA demonstrates up to 97.3% latency reduction while maintaining accuracy comparable to gate-based approaches when evaluated on real quantum devices. Finally, we explore error mitigation techniques for VQAs at the pulse level. We develop a fidelity estimator based on reversed pulses, that enables randomized benchmarking of parametric pulses. This estimator compares the final state obtained after applying a sequence of pulses followed by their reversed counterparts to the initial state, using the probability of successful trials as a proxy for fidelity. Furthermore, we adapt the zero-noise extrapolation (ZNE) technique to the pulse level, enabling the error mitigation for quantum pulses. Applied to VQE tasks for H2 and HeH+ molecules, pulse-level ZNE reduces the deviation from ideal expectation values by an average of 54.1%. The techniques developed in this thesis advance the efficiency and practicality of VQAs on near-term quantum devices. The introduced frameworks, AccQOC and EPOC, provide efficient pulse optimization, while TopGen and NAPA can construct hardware-efficient ansatz. Besides, the pulse-level error mitigation techniques presented in this thesis improve the resilience of VQAs against the inherent noise and imperfections of NISQ devices. Together, these contributions help unlock the full potential of quantum computing and realize practical quantum advantages in the near future.</p>