Autism Spectrum Disorders, hereafter referred to as autism, emerge early and persist throughout life, contributing significantly to global years lived with disability. Typically, an autism diagnosis depends on clinical assessments by highly trained professionals. This high resource demand poses a challenge in resource-limited areas where skilled personnel are scarce and awareness of neurodevelopmental disorder symptoms is low. We have developed and tested a novel app, START, that can be administered by non-specialists to assess several domains of the autistic phenotype (social, sensory, motor functioning) through direct observation and parent report. N=131 children (2-7 years old; 48 autistic, 43 intellectually disabled, and 40 typically developing) from low-resource settings in the Delhi-NCR region, India were assessed using START in home settings by non-specialist health workers. We observed a consistent pattern of differences between typically and atypically developing children in all three domains assessed. The two groups of children with neurodevelopmental disorders manifested lower social preference, higher sensory sensitivity, and lower fine-motor accuracy compared to their typically developing counterparts. Parent-report further distinguished autistic from non-autistic children. Machine-learning analysis combining all START-derived measures demonstrated 78% classification accuracy for the three groups (ASD, ID, TD). Qualitative analysis of the interviews with health workers and families (N= 15) of the participants suggest high acceptability and feasibility of the app. These results provide proof of principle for START, and demonstrate the potential of a scalable, mobile tool for assessing neurodevelopmental disorders in low-resource settings.