Machado-Joseph disease, also known as Spinocerebellar ataxia type 3 (MJD/SCA3), is a fatal autosomal dominant hereditary ataxia characterized by cerebellar ataxia resulting from the abnormal expansion of CAG repeats in exon 10 of the ATXN3 gene. Presently, there is no effective treatment for SCA3. Small interfering RNAs (siRNAs) are emerging as potential therapeutic strategies to specifically target the disease-causing mutant ATXN3 (mATXN3) protein. However, the delivery efficiency of siRNAs remains a major obstacle for clinical application, particularly in brain disorders. This study aimed to develop a synthetic biology strategy to reprogram the host liver as a tissue chassis to induce and deliver in vivo self-assembled siRNAs (IVSA-siRNAs) to target the ATXN3 gene. A synthetic construct directed by a cytomegalovirus promoter was designed to encode a neuron-targeting rabies virus glycoprotein tag and mATXN3-siRNA. After intravenous injection, the synthetic construct was taken up by mouse livers, which were then reprogrammed to enable the self-assembly, production, and secretion of small extracellular vesicles (sEVs) encapsulating mATXN3-siRNA. The sEV-encapsulated mATXN3-siRNA was further transported through the endogenous circulating system of sEVs, crossing the blood-brain barrier and reaching the cerebellar cortex and spinal cerebellar tract, where they silenced the ATXN3 gene. Treatment with the synthetic construct for 8 or 12 weeks led to significant improvements in motor balance ability and reduction of cerebellar atrophy in YACMJD84.2 transgenic mice. The number of Purkinje cells in the cerebellar cortex was significantly increased, and the loss of myelin basic protein was reduced. Moreover, the quantity of neurotoxic nuclear inclusion bodies and the expression of glial fibrillary acidic protein, which promotes neuroinflammation in activated astrocytes, were decreased significantly. The synthetic construct facilitated the generation and delivery of IVSA-siRNA to the cerebellar cortex and spinal cerebellar tract, thereby inhibiting the expression of mATXN3 protein. This treatment successfully addressed motor impairments, alleviated neuropathological phenotypes, and mitigated neuroinflammation in YACMJD84.2 transgenic mice. Our strategy effectively overcomes the primary challenges associated with siRNA therapy for cerebellar ataxia, offering a promising avenue for future clinical treatments.