Customer delivery / Case 01
Huiwan / WePlay — Voice AI in Production
Outcome & decision
Led solution design and customer delivery for a production voice AI companion. Aligned customer requirements, integration testing, and engineering work across the real-time audio and speech-to-response pipeline.
- My responsibility
- Solution design and customer delivery coordination at ZEGO
- Project scope
- Production voice-AI delivery; later multi-user expansion work
Evidence basis. Delivery scope is documented in my career records. Customer records remain private; no audited latency or adoption metric is presented here.
Customer problem & constraints
Huiwan needed an AI companion for social voice experiences. Real-time communication (RTC), automatic speech recognition (ASR), model responses, and text-to-speech (TTS) had to behave as one responsive conversation.
Integration changes also had to account for an operating service: SDK behavior, API compatibility, concurrent use, and customer expectations could not be validated in isolation.
My role & team responsibilities
As the senior pre-sales lead, I owned solution design and customer delivery coordination: requirements, acceptance criteria, issue priorities, and communication between the customer and delivery teams.
I worked with product, engineering, QA, sales, and delivery colleagues on integration and latency analysis. Engineering implementation and performance changes were team work; my responsibility was to connect the technical decisions to customer requirements and delivery readiness.
Key decisions & execution
I treated the full RTC → ASR → LLM/RAG → TTS path as the unit of integration. This made it possible to discuss response behavior across service boundaries, rather than evaluate each component only on its own.
I coordinated API upgrade and compatibility planning for existing traffic. Daily delivery tracking and a shared issue process kept owners, priorities, and acceptance criteria visible as customer feedback reached product and engineering.
Delivery & validation
The voice AI companion reached production. I coordinated test-case reviews, server load testing, SDK regression, and production monitoring to support the delivery process.
Later work addressed multi-user conversations and background music. That expansion is separate from the established production milestone; its launch status is not asserted here.
Evidence & limitations
Career records support the production-delivery account, but drafts differ on timing and scenario scope. An approximate latency reduction is recorded without its baseline, percentile, or measurement conditions, so it is excluded from the headline.
The case establishes solution and delivery responsibility. It does not establish a response-time service-level objective, concurrency level, user adoption, or incremental revenue.