AI comic drama production
Technical lead, connecting script understanding, character assets and storyboard generation with quality checks and recovery.
≈18% → 9.53% Issue rate on a fixed evaluation set← BackPERSON / BEHIND THE FOUR STATES
KENNY WU / SHENZHEN / AI ENGINEEROne person,
One person,
behind the lenses.
I’m Kenny Wu, an AI engineer in Shenzhen. My work spans NLP, information retrieval, multimodal understanding and content generation. I care about turning advances in models into stable, useful products.
Contact me ↗
ONE PERSON / FOUR STATES
EXPERIENCE / SELECTED
The work so far
Short-drama editing and character consistency
Cross-episode character alignment and highlight retrieval keep automated editing focused on the same protagonist.
1285s → 24s Identity disambiguation on one 10-episode sampleMore experience, research and everyday life
Experience
AI comics & storyboard generation
Technical lead
Helped shape the system architecture and core workflow, connecting script understanding, asset extraction, and storyboard generation. Led quality evaluation and acceptance criteria, with recovery mechanisms for long-running generation tasks.
Fixed evaluation set: action-overload and shot-scale issues fell from about 18% to 9.53%. One 20-episode run completed despite intermediate failures.
Work & outcomes
- Decomposed scripts into structured multi-shot storyboards, enforcing independent quality grading and bounded-cost auto-correction.
- On a fixed evaluation set, action-overload and shot-scale issues fell from about 18% to 9.53%—a relative reduction of about 47%.
- Added circuit breaking, concurrency control, and bounded recovery; a 20-episode run completed despite intermediate failures.
Short-drama editing & character identity
Workstream lead
Solved character drift across episodic short dramas via cross-episode identity tracking, entity alignment, and highlight retrieval, keeping automated cuts locked onto the main cast.
One 10-episode sample: identity disambiguation fell from 1,285s to 24s; understanding orchestration fell from 23 minutes to about 5–6 minutes.
Work & outcomes
- Improved data services, concurrent scheduling, and cache reuse to support the editing pipeline.
- On one 10-episode sample, identity disambiguation fell from 1,285 to 24 seconds; understanding orchestration fell from 23 minutes to about 5–6 minutes.
E-commerce copy style adaptation
Core algorithm engineering
Engineered an end-to-end stylized copywriting pipeline for digital human livestreams, spanning automated dataset synthesis, supervised fine-tuning (Qwen2.5-7B), and multi-metric evaluation.
Fine-tuned Qwen2.5-7B for five copy styles.
Work & outcomes
- Designed a source-text → style-profile → rewrite dataset workflow and multidimensional automatic evaluation.
- Fine-tuned Qwen2.5-7B for five copy styles, using structured style information to stabilize training.
Enterprise RAG, safety & intent recognition
Core algorithm engineering
Worked on enterprise document Q&A, internal LLM safety review, and voice-bot intent recognition, covering retrieval, generation, and interaction quality.
Internal tests reported about 90% recall and Q&A accuracy.
Work & outcomes
- Architected an enterprise RAG system with hierarchical document trees and hybrid retrieval, achieving ~90% recall and answer accuracy in internal benchmarks.
- Combined semantic matching and rules for content review, supporting the internal model registration process.
- Used Aho–Corasick matching and inverted indexes for intent recognition across scenarios.
Intelligent search & conversation analysis
NLP algorithm trainee
Worked on intelligent search, Text2SQL, and social-media support conversations, covering query understanding, similar-question grouping, and discovery of unanswered questions.
Work & outcomes
- Generated SQL templates from database schemas and plain-language explanations for business review.
- Used contrastive sentence embeddings and hierarchical density clustering to expand question libraries and support conversation reporting.
Research & publications
Semantic matching based legal information retrieval system for COVID-19 pandemic
Junlin Zhu, Jiaye Wu, Xudong Luo, Jie Liu
Alignment-based graph network for judicial examination task
Jiaye Wu, Xudong Luo
Awards
China AI and Law Challenge (CAIL)
- 2nd Prize · Legal Case Retrieval
- 3rd Prize · Information Extraction
- 3rd Prize · Judicial Examination
With Prof. Zhiyuan Liu.
Guangxi Collegiate AI Design Competition
1st Prize
Subway Passenger Flow Prediction
Guangxi Collegiate AI Design Competition
2nd Prize
CSI Index Prediction
Three questions I return to
What deserves to become lasting context? ↗
Away from the screen
Some music, a few pages, a long walk.
A change of pace for my attention, and room for questions I haven't worked out yet.
Professional links and contact
Google Scholar ↗ · GitHub ↗ · CV ↗
jdlow@live.cn ↗