CV

Work Experience

Tencent Ads · Algorithm Researcher · Shanghai (2018 – present)

• Generative Recommendation (2025.8 – present)
  • Applying generative retrieval to the ads recommendation pipeline.
• AIGC Creative Generation (2023 – 2025.8)
  • Built multi-model AIGC code frameworks covering image generation, image understanding, and video editing; supported online serving and API for dozens of models.
  • Product poster generation: Background generation model + adaptive marketing element rendering, producing diverse high-quality posters for e-commerce advertisers while preserving the product subject.
  • Stylized image synthesis: Learned client creative styles to build personalized image generation models for high-demand advertisers.
  • Promoted AIGC generation features into the final-stage ranking model to improve AI creative delivery.
  • Monitored and optimized AI creative audit pass rate, fingerprint duplication rate, and automated trigger volume.
• Retrieval Systems (2021 – 2022)
  • Led next-generation retrieval architecture: collaborated to sunset targeting-based retrieval, establishing an LTR-centric framework with complementary models and precision targeting filters.
  • LTR model iterations: negative sampling strategies, sample weighting, loss function optimization.
  • Multi-interest model iteration: introduced long-sequence feature modeling.
  • Responsible experiments contributed to cumulative GMV lift of 2.2%; core participation contributed 2.0%.
• Contract Advertising (2018 – 2020)
  • Participated in building a distributed retrieval system for contract ad inventory estimation; incorporated freshness, frequency capping, and inventory contention factors to improve overall sellable inventory rate.

---

TouchPal (NYSE: COOL) · Algorithm Engineer (2014 – 2017)

• Multilingual language modeling: Built ngram LM pipeline (crawling → corpus cleaning → n-gram counting → smoothing); later upgraded to LSTM deep model. Achieved candidate word prediction within 50ms on mobile devices.
• Key-press error correction: Designed a correction mechanism combining key press position heatmaps with the language model. Outperformed Nuance (then the leading international IME) in evaluations by multiple OEM partners.

Skills

• Machine Learning: Deep learning, LTR, multi-interest modeling, generative retrieval, sequence-to-sequence
• AIGC: Diffusion models, image generation, video editing, style transfer, model serving
• Systems: Distributed retrieval, online inference serving, large-scale feature engineering
• Languages: Python, C++
• Competitive Programming: ACM-ICPC, Codeforces (handle: _LT_zyc), AtCoder (handle: zyczyc)