What Baidu Unveiled at QCon 2021: Key Takeaways from 7 Cutting‑Edge Sessions

01. Unified Quality & Efficiency Delivery for Baidu Feed Recommendation

Speaker: Zhang Sainan, Senior Test Engineer, User Quality & Efficiency Department, Baidu

Abstract: As Baidu's feed recommendation business expands, the system becomes massive and complex, containing hundreds of modules, strategies, and models. Ensuring both quality and efficiency during rapid, large‑scale iterations poses a major QA challenge. Traditional testing adds repetitive steps and low‑level checks. In the intelligent era, a data‑driven, algorithm‑guided delivery pipeline enables lightweight, high‑efficiency operation, supporting fast business iteration while maintaining quality.

Audience Benefits:

Learn a systematic approach to guarantee quality and efficiency in large‑scale recommendation systems.

Understand how data and algorithms can drive a lightweight, high‑throughput delivery workflow.

02. Software Engineering Capability Talk

Speaker: Zhang Miao, Ph.D., Cloud Architecture Engineer, Baidu Intelligent Cloud

Abstract: With industry upgrades and rising labor costs, software engineering capability has attracted widespread attention. This talk explains why engineering capability matters, defines the concept, and shares practical ways to strengthen it.

Audience Benefits:

Deepen understanding of engineering capability.

Learn concrete methods to improve personal and team engineering skills.

03. AIOps Best Practices in Fault‑Management Scenarios

Speaker: Chen Yun, Senior R&D Engineer, Baidu Fault‑Management Team

Abstract: Fault management is critical for high availability, yet traditional troubleshooting relies on engineers manually analysing real‑time monitoring data, which is unreliable at scale. Even small distributed systems generate massive metrics, making comprehensive analysis difficult. Baidu explores algorithmic solutions for anomaly detection on key metrics, automated traffic switching for single‑datacenter failures, data‑driven fault diagnosis, and predictive fault prevention, turning firefighting into fire‑prevention.

Audience Benefits:

Understand Baidu's fault‑management experience and challenges.

Learn intelligent operations algorithms for fault detection, stop‑loss, diagnosis, and prediction.

Gain insight into practical case studies of AIOps deployment.

04. Real‑Time Analytics with PB‑Level MPP Database Apache Doris

Speaker: Chen Mingyu, Senior R&D Engineer, Baidu

Abstract: Apache Doris (Incubating) is Baidu's open‑source version of the Palo analytical database, contributed to Apache in 2018. It powers PB‑scale data with low‑latency, interactive exploration and supports high‑concurrency online reporting. The talk covers Doris's architecture, core features, real‑time stream ingestion, and recent advances for online data scenarios.

Audience Benefits:

Grasp the design philosophy behind Doris's MPP architecture.

Identify typical application scenarios and technical solutions using Doris.

05. Service Mesh at Baidu’s Hundred‑Billion‑Level Production Scale

Speaker: Chen Peng, Cloud‑Native Technology Expert, Baidu

Abstract: Service Mesh has quickly become the communication standard in cloud‑native environments, yet leading solutions like Istio still face limitations in heterogeneous infrastructure support, private protocol handling, complex traffic routing, productization, performance, and reliability. This session presents Baidu's real‑world deployment of Service Mesh across Feed, Mobile Baidu, and Baidu Maps, highlighting pain points and practical experiences.

Audience Benefits:

Learn how to safely and smoothly roll out Service Mesh at massive production scale.

06. Large‑Scale Service Governance and Fault Prevention

Speaker: Zhen Zhen, Senior R&D Engineer, Baidu Search Architecture

Abstract: In a massive micro‑service architecture, faults occur frequently. Baidu shares a series of technical solutions for high availability, including traffic‑scheduling frameworks to improve system resilience, precise stop‑loss mechanisms to minimize loss, and comprehensive observability and automated analysis to achieve white‑box search systems.

Audience Benefits:

Understand the availability challenges of large‑scale micro‑service systems.

Explore extreme optimization strategies for communication frameworks.

Learn design patterns for precise stop‑loss and innovative observability.

07. Deep‑Learning Technology Insights and Applications at Baidu

Speaker: Yan Chunwei, Senior Engineer, Deep Learning Platform, Baidu

Abstract: The session introduces Paddle Lite, a lightweight inference engine for mobile and IoT devices, detailing its architecture, design decisions, and typical use cases. It also covers the Paddle core framework, the open‑source model library (image classification, object detection, NLP), and the AI development dual‑platform model, highlighting zero‑threshold development, full‑feature modeling, and core technologies.

Audience Benefits:

Understand the architecture and deployment scenarios of Paddle Lite for mobile inference.

Learn about Paddle’s core framework upgrades and open‑source model offerings.

Gain insight into Baidu’s AI development platform, auto‑DL capabilities, and enterprise‑grade AI solutions.