This study presents a neural-network–driven digital platform that integrates multimodal urban data to manage social and labor relations in megacities. Addressing the persistent mismatch between labor demand and supply in urban centers, the platform integrates multimodal data sources—including municipal job registries, payroll ledgers, social-media streams, and IoT mobility sensors—into unified embeddings of occupations, skills, and locations. Using a real-time data-ingestion pipeline (≤5-min latency) we fused municipal job‐registry feeds, payroll ledgers, social-media streams, and IoT mobility sensors into unified embeddings for occupations, skills, and locations. The hybrid forecasting core—an LSTM time-series channel cross-attended with a graph neural network (GNN) of worker–skill relations—achieved a mean absolute percentage error (MAPE) of 6.8% when predicting quarterly labor-demand gaps in Ho Chi Minh City (target recruitment 79 000–84 000 workers for Q1 2025) and across Russia’s federal districts (regional unemployment ranging from 1.7% to 8.3%). Real-time alerting (<1 s) enabled employment-service overload detection during peak registration bursts—e.g., Hanoi’s creation of 124 920 jobs in H1 2024 (75.7% of its annual goal). A Deep-Q-Learning decision-support layer raised the return on social programs by ≥15% while fairness constraints plus XAI (LIME, SHAP) eliminated measurable disparate impact (Δ → 0). These results demonstrate how advanced neural architectures can simultaneously improve forecast accuracy, policy ROI, and algorithmic accountability in rapidly evolving urban labor markets. Overall, the platform offers a scalable solution for balancing workforce supply and demand, supporting inclusive policies, and enhancing resilience in megacity labor markets.