Collagen is widely used as a biomaterial in tissue engineering and regenerative medicine. However, growing concerns over animal-derived sources have motivated the development of synthetic human collagen to minimize associated risks. Here, we present a synthetic biology strategy for microbial production of human type I collagen (hCOL1) fragment, integrating machine learning with biofoundry-driven workflows. Using an ESM-2 protein language model combined with biofoundry-based strain development, context-dependent synthetic signal peptides (SP) were designed and evaluated. This approach achieved the production and secretion of hCOL1 fragments in Corynebacterium glutamicum as an industrial host at 24.3 ± 3 mg/L after the third round of strain engineering, representing a 4.9-fold increase compared with the native SP in the first round. Next, scalable hCOL1 fragment production was demonstrated in 5-L fermenters, achieving titers of 857 ± 75 mg/L in fed-batch cultivations using glucose. To validate its potential as a tissue engineering biomaterial, the collagen-based biocomposite scaffolds were fabricated and used to culture human adipose-derived stem cells. The engineered matrices significantly enhanced proliferation and stemness marker expression, surpassing conventional animal-derived collagens. Active learning for context- and host-dependent SP design can facilitate the secretion of not only collagen but also other proteins of interest.