SMARTFOOD: An LLM-Driven Ubiquitous Mobile App for Context-Aware Restaurant Recommendation

We introduce SMARTFOOD, a context-aware mobile app and rich user profiles (dietary restrictions, allergies, taste preferences, health status) to deliver personalized, restaurant-level food recommendations that promote healthier choices. By combining nutritional insights from the Finnish Fineli database with a fine-tuned LLM, SMARTFOOD not only suggests venues and dishes that fit each user’s unique context – whether a quick office lunch or a family dinner – but also highlights options lower in calories. In performance tests on three Android configurations, app startup time ranged from 2.2 s to 3.4 s, lunch?menu retrieval from 6.5 s to 7.5 s, and screen loads under 0.8 s. Average memory usage was 370 MB (emulators) to 470 MB (S22 Ultra) with peak CPU under 55%, indicating robust cross-device performance. In a small unmoderated user study (N=5), all participants achieved 100% task completion (account setup, menu fetching/feedback, profile updates) with low error rates (<4.1%) and mean times per task under 3.7 minutes. Usability scales were high (SUS = 83.6 ± 8, UMUX-LITE = 80 ± 6), confirming the app's ease of use and effectiveness. Our source code is available at https://github.com/ValtteriJokisaari/smartfood. © 2025 Owner/Author.