Backend
直接向此候选人发送职位邀请
Results-driven Backend Software Engineer with 5+ years of experience architecting high-concurrency, data-intensive systems for millions of daily active users. Proven track record in orchestrating complex microservices, implementing strict anti-overselling transaction mechanisms, and designing resilient traffic gateways. Adept at driving systemic stability (99.99% SLA) and infrastructure FinOps optimization in large-scale local services platforms.
Meituan | Top Performer (2025H1/2025H2) Shanghai, ChinaRole Context: Tech Lead for the C-end discovery and booking funnel, orchestrating high-concurrency traffic gateways and dynamic rendering systems across 50+ industry verticals.
– Project: Millions-Scale UGC AI Feature Scheduling & Commercial Insight Engine
∗ Architected a Lambda-based asynchronous scheduling pipeline processing 10M+ daily reviews, bridging massive data surges and strict LLM API quotas (500 QPS limit).
∗ Engineered a resilient Kafka consumer gateway utilizing dual-window micro-batching. Implemented a Dead Letter Queue (DLQ) to isolate LLM-crashing “poison pills” and a SkipList offset tracker to guarantee at-least-once semantics.
∗ Developed a commercial insight engine via two-stage Spark salting and Milvus vector clustering to extract merchant highlights from unstructured reviews. Deployed a Geo-Spatial TF-IDF scoring algorithm, driving a 3% CTR uplift.
– Project: Read-Intensive Traffic Aggregation Gateway for LBS Booking Shelf
∗ Architected a high-concurrency aggregation layer utilizing a consolidated Redis view cache to bridge the gap between 12K QPS discrete LBS read traffic and an 8K QPS downstream inventory hard quota.
∗ Optimized payload physics to bypass downstream ETag dependencies. Engineered Protobuf serialization and Zstd compression to shrink 1MB full-view payloads by 80%, sustaining P99 latency under 120ms without overloading NIC bandwidth limits.
∗ Engineered an MQ-depth-driven adaptive TTL backpressure mechanism to prevent cache stampedes. Implemented an atomic in-flight Redis lock (SETNX) with active consumer-side release to strictly eliminate update amplification, ensuring a 99.99% SLA.
– Project: High-Concurrency SDUI BFF Engine & DAG Execution Architecture
∗ Led the architectural evolution of the SDUI-based BFF engine for Meituan’s C-end funnel. Decoupled frontend rendering from backend logic via Zookeeper-managed schemas, slashing A/B testing and UI iteration cycles from weeks to minutes.
∗ Refactored the routing architecture to handle explosive UI combinations across 50+ verticals and hundreds of concurrent A/B experiments. Implemented AOT compilation for structural DAGs, reducing gateway CPU overhead by 35%.
∗ Pioneered a field-level DAG concurrency model with Partial Degradation. Automatically isolated non-core downstream timeouts while preserving core transactional fields, preventing cascading failures and sustaining a 99.99% rendering rate.
Meituan | Top Performer (2022H1) Shanghai, ChinaRole Context: Independent Lead for the Reservation & Booking module, managing core inventory logic and orchestrating integration with technicians, rooms, and order management systems for 20K+ merchants.
– Project: High-Concurrency SaaS Booking Engine & Anti-Overselling Architecture
∗ Architected a resilient booking engine for 20K+ merchants. Resolved a 300 TPS legacy bottleneck by implementing Redisson distributed locks with lexicographical sorting to prevent deadlocks, boosting core API throughput to 2.5K TPS.
∗ Engineered strict transaction boundaries (post-commit lock releases) to systematically prevent phantom overselling. Designed a graceful degradation strategy via global rate limiting and merchant-tier prioritization during Redis outages.
∗ Achieved cross-service eventual consistency via a Transactional Outbox pattern. Combined Kafka fast-path delivery with a polling worker for delayed records, reducing manual reconciliation efforts by 95% and maintaining 18 months of zero-overselling incidents.
