Backend systems · AI infrastructure
Yupeng Ma
Senior Backend Engineer Python · Go · AI Infrastructure
8 years of experience building and shipping production backend systems.
ByteDance · iQIYI · Production AI Infrastructure
View Projects View Resume GitHub ↗ Contact Me Open to remote backend opportunities outside China.
Highlights
A quick read on where I add value.
8 Years Production Backend Engineering
ByteDance · iQIYI Large-scale production systems
Python · Go · AI Backend & Infrastructure Featured projects
Systems and tools, explained through engineering decisions. A quick view of the problems, responsibilities, and results behind selected backend work. Feature-driven AI delivery loop
01 Product Source of truth 02 Feature Bounded scope 03 AI agent Implementation 04 Review Human gate 05 Ship Accepted change product → feature feature → agent agent → review review → ship ship → productnext loop A product specification becomes a bounded feature, an AI agent implements it, human review verifies the result, and the accepted change ships before the loop continues. A feature-driven workflow for shipping software with AI coding agents and explicit human review.
Real-time AI service flow
01 Client Headset product 02 API FastAPI control 03 Streaming WebSocket session 04 AI model Voice synthesis 05 Logs ELK observability client → api api → stream stream → model api → logsobserve A product client reaches a FastAPI control boundary, opens a streaming session, and coordinates with an AI model. Operational logs branch from the backend boundary. Production APIs and real-time streaming for an AI voice feature.
Multi-vendor allocation flow
01 Traffic Forecast demand 02 Allocator Decision support 03 Vendor A CDN provider 04 Vendor B CDN provider 05 Monitoring Performance context inputs → allocator allocator → vendor-a allocator → vendor-b monitoring → allocatorfeedback Planning inputs enter an allocation engine, which supports traffic decisions across multiple vendors while monitoring provides operational feedback. Traffic allocation across CDN vendors using forecast and performance signals.
View all projects →