A major challenge in industrial biomanufacturing of therapeutic alkaloids is the slow process of biocatalyst engineering. Amaryllidaceae alkaloids, such as galantamine, are complex plant secondary metabolites with therapeutic value. Due to their difficult synthesis, they are often extracted from daffodils. This study presents an efficient biosensor-machine learning technology stack for biocatalyst development, applied to engineer an Amaryllidaceae enzyme in E. coli. Directed evolution was used to develop a highly sensitive biosensor for the key Amaryllidaceae alkaloid branchpoint 4'-O-methylnorbelladine. A structure-based residual neural network (MutComputeX) was developed to generate activity-enriched variants of a plant methyltransferase, which were rapidly screened with the biosensor. Functional enzyme variants were identified that yielded a 60% improvement in product titer, 2-fold higher catalytic activity, and 3-fold lower off-product formation. A crystal structure elucidates the mechanism behind key beneficial mutations. Amaryllidoideae alkaloids have therapeutic promise, including anticancer, fungicidal, antiviral, and acetylcholinesterase inhibition properties. Galantamine, a selective and reversible acetylcholinesterase inhibitor, is a licensed treatment for Alzheimer's disease. Due to its challenging synthesis, global supplies rely on isolating low quantities from daffodils, resulting in an expensive and environmentally-dependent supply chain. To improve galantamine production, agricultural techniques are being tested to boost daffodil yields. A promising alternative to plant extraction is microbial fermentation. Recent studies have reconstituted long plant pathways into microbial hosts for the production of therapeutic alkaloids. While the complete biosynthetic pathway for any AA with therapeutic value has not yet been elucidated, recent studies have characterized early pathway enzymes responsible for the biosynthesis of 4'-O-Methylnorbelladine, the last common intermediate before AA pathway branches diverge. Semi-synthetic methods using characterized enzymes have been proposed to generate advanced intermediates. The industrial application of such pathways could be greatly accelerated by augmenting high-throughput screens with genetic biosensors and using machine learning to guide protein design, yielding enzymes and pathways with improved stability and activity. Here, we synergize the development of custom biosensors with machine learning (ML)-guided protein design to improve microbial fermentation of the branchpoint AA 4'-O-methylnorbelladine (4NB). A generalist transcription factor, RamR, was evolved into a highly sensitive biosensor for 4NB that precisely discriminates against the non-methylated precursor norbelladine. The biosensor was then used to monitor the activity of norbelladine 4'-O-methyltransferase (Nb4OMT) from the daffodil Narcissus pseudonA major challenge in industrial biomanufacturing of therapeutic alkaloids is the slow process of biocatalyst engineering. Amaryllidaceae alkaloids, such as galantamine, are complex plant secondary metabolites with therapeutic value. Due to their difficult synthesis, they are often extracted from daffodils. This study presents an efficient biosensor-machine learning technology stack for biocatalyst development, applied to engineer an Amaryllidaceae enzyme in E. coli. Directed evolution was used to develop a highly sensitive biosensor for the key Amaryllidaceae alkaloid branchpoint 4'-O-methylnorbelladine. A structure-based residual neural network (MutComputeX) was developed to generate activity-enriched variants of a plant methyltransferase, which were rapidly screened with the biosensor. Functional enzyme variants were identified that yielded a 60% improvement in product titer, 2-fold higher catalytic activity, and 3-fold lower off-product formation. A crystal structure elucidates the mechanism behind key beneficial mutations. Amaryllidoideae alkaloids have therapeutic promise, including anticancer, fungicidal, antiviral, and acetylcholinesterase inhibition properties. Galantamine, a selective and reversible acetylcholinesterase inhibitor, is a licensed treatment for Alzheimer's disease. Due to its challenging synthesis, global supplies rely on isolating low quantities from daffodils, resulting in an expensive and environmentally-dependent supply chain. To improve galantamine production, agricultural techniques are being tested to boost daffodil yields. A promising alternative to plant extraction is microbial fermentation. Recent studies have reconstituted long plant pathways into microbial hosts for the production of therapeutic alkaloids. While the complete biosynthetic pathway for any AA with therapeutic value has not yet been elucidated, recent studies have characterized early pathway enzymes responsible for the biosynthesis of 4'-O-Methylnorbelladine, the last common intermediate before AA pathway branches diverge. Semi-synthetic methods using characterized enzymes have been proposed to generate advanced intermediates. The industrial application of such pathways could be greatly accelerated by augmenting high-throughput screens with genetic biosensors and using machine learning to guide protein design, yielding enzymes and pathways with improved stability and activity. Here, we synergize the development of custom biosensors with machine learning (ML)-guided protein design to improve microbial fermentation of the branchpoint AA 4'-O-methylnorbelladine (4NB). A generalist transcription factor, RamR, was evolved into a highly sensitive biosensor for 4NB that precisely discriminates against the non-methylated precursor norbelladine. The biosensor was then used to monitor the activity of norbelladine 4'-O-methyltransferase (Nb4OMT) from the daffodil Narcissus pseudon