Big Data Architecture and Solutions for Financial Scenarios: MMR, Honghu Data Lake, and Yichuang Model Training Platform

The financial sector faces increasing demands for fine‑grained control over the entire data processing pipeline, while users desire low‑threshold, fast access to data and companies need to capture operational experience.

Solution 1: MMR Cloud‑Based Big Data Architecture – Built on Baidu Cloud products, the architecture extends the standard stack with five layers: access, table‑control, compute, virtual storage, and physical storage. The user layer integrates employee management to tag operations with personal identities, enabling precise accountability. The table‑control layer provides field‑level permission for Hive tables, allowing selective sharing of sensitive columns. The compute layer leverages Baidu Cloud resources and introduces a virtual management layer to isolate unstructured data while supporting directory‑level permissions and IP‑based access controls. Unified client tools smooth the transition from legacy Baidu architecture to the open‑source cloud stack, and intelligent scheduling with windowed execution and heartbeat agents ensures high availability and rapid fault diagnosis.

Solution 2: Honghu Data‑Lake Management and Analysis Platform – Aimed at strategy analysts, the platform lowers the cognitive barrier through unified metadata aggregation, topic‑domain construction with intelligent recommendation, strict data‑quality monitoring, and comprehensive permission, masking, and encryption controls. It also offers a visual, drag‑and‑drop batch & streaming development environment supporting Hive, Spark, Flink, GP, and Shell, as well as a one‑click data‑exchange service and API for downstream system integration.

Solution 3: Yichuang Model‑Training Monitoring and Evaluation System – Provides end‑to‑end one‑click model training, standardizing code, environment, sample and feature libraries, and ensuring consistency between offline and online feature stores. The platform includes a plugin‑based evaluation framework that unifies metrics across models, facilitating consistent assessment and deployment.

The session concludes with a forward‑looking discussion on embracing cloud‑native, lake‑warehouse integration, and further unlocking data value, followed by a Q&A where the consistency between online and offline features is explained.

Speaker: Zhao Hui, Architect at Du Xiaoman Financial; Organizer: Xiamen University Jiageng College; Platform: DataFunTalk.