Accurate dietary assessment in non-Western populations is hindered not only by recall bias but also by the limited representation of local foods in international reference databases. When standardised tools fail to capture regional dietary practices, the quality of dietary assessments is potentially compromised. This cross-sectional study evaluated the accuracy of photograph-assisted 24-h recalls (24hr) analysed using the United States Department of Agriculture (USDA) Automated Self-Administered 24-Hour Dietary Assessment Tool (ASA24) system (based on Food and Nutrient Database for Dietary Studies (FNDDS) 2017–2018) or a customised database enriched with local foods and international measurement units (DbF). Twenty Colombian adults photographed and weighed foods consumed over two non-consecutive days. A dietitian conducted 24hr interviews using the multiple-pass method based on the photographs. Intake estimated with ASA24 and DbF was compared against weighed food records as reference. Blood and urine samples were collected to assess nutrient biomarkers. Accuracy was evaluated using correlation analyses and Bland–Altman plots, while usual intake was estimated using PC-SIDE. Misreporting was similar between ASA24 and DbF (15 %). However, DbF improved accuracy, showing stronger correlations with weighed foods for calories and macronutrients (ρ > 0·90) and narrower limits of agreement for energy intake compared with ASA24 (−364 to 256 v. −771 to 909 calories, respectively). DbF allowed estimation of twenty-eight nutrients compared with twenty using ASA24 and showed stronger correlations for micronutrient intake estimates (r 0·86–0·96 v. 0·69–0·96; all P < 0·05) and urinary biomarkers (r > 0·6, P < 0·0001). These findings demonstrate that integrating locally contextualised food databases with technology-assisted recalls improves the quality of dietary assessment in underrepresented populations and provides a scalable model to enhance nutritional epidemiology and health monitoring worldwide.