Define what good means first.
Turn quality concerns into measurable criteria. Use a fixed evaluation set to check whether changes help.
KENNY WU · AI ENGINEER · SHENZHEN
From AI capability
to reliable delivery.
I’m Kenny, an AI application and systems engineer.
I bring language, retrieval, and multimodal models into real workflows.
01 / SELECTED DELIVERY
Professional work · Roles & outcomes· Technical lead
AI comics generation
Beyond generation,
quality checks and recovery matter.
Connected script understanding, asset extraction, and storyboard generation. Led quality evaluation and acceptance criteria.
See responsibilities and outcomes ↗From script to evaluated storyboardsWorkflow diagram
- 01 / UNDERSTANDUnderstand scriptsIdentify characters and plot
- 02 / ORGANIZEExtract assetsOrganize characters and scenes
- 03 / GENERATEGenerate shotsStructured shot decomposition
Independent checks→Bounded correction→Acceptance
Runtime safeguards: circuit breaking · concurrency control · bounded recovery
Issue rate · Fixed evaluation set
≈18%→9.53%
Action overload & shot-scale issues · ≈47% relative reduction
One real-world run
20 episodes
Completed despite intermediate failures
2024 / RETRIEVAL & ANSWERING
Enterprise document Q&A
Combined paragraph hierarchy, multi-path retrieval, and dynamic thresholds into a document Q&A workflow.
See this work ↗Document structureMulti-path retrievalAnswer evaluation
≈90%
Internal tests
Recall and Q&A accuracy
02 / ENGINEERING JUDGMENT
How I move the work forward
Build failure into the workflow.
Make long tasks recoverable and bound correction costs, so one failure does not stall delivery.
Connect models to real work.
Check the workflow from retrieval and generation to service orchestration, bringing results to acceptance.
03 / OPEN-SOURCE WORK
Ideas become useful tools, too.
04 / FIELD NOTES
Field notes
LET’S WORK TOGETHER
Have an AI project to move forward?
Let’s discuss the technical approach, engineering delivery, quality, and reliability.