Cai Pengyang

As the director of pricing engineering team, I’m collaborating with other teams including Science, Analytics, Product to assure the sharpness goal of our retail business. We are developing the pricing intelligence solutions to guarantee our price competitiveness among competitors in South Korea. We developed the automapping system and introduced different intelligent strategies to improve the business performance, such as SBC(search based crawling), VBS(vector based search) sparse retrieval and graph driven retrieval.

Meanwhile, we also finished the legacy systems deprecation successfully within 3 months and resolved the tech debts which last for more than 5 years. Now the whole team is focusing on improving the intelligence, scalability of the whole system from end to end leveraging AI technologies including traditional AI technologies like NLP, CV as well as LLM and LVLM.

Highlights

• Resolved 5 - year tech debts within my first two quarters Our legacy system was built 5+ years ago and was not maintained well. It was using outdated big data solutions and the performance was not good. And the solution also couldn't scale as business grew. Due to too many contributors and too many business requirements, the codebase was a mess and it made the system hard to maintain and supporting any new feature was a nightmare. I led the team to redesign the system and rebuild the core components. We learned from the search & recommendation system architecture and came up with near real-time backend system solution and eventually we built a high performance, scalable, and maintainable system. The throughput of the system was improved by 3X and the cost dropped by 85%. And supporting new features was much easier than before. The architecture design, module relation, service framework, and coding style were well defined.

As the manager of the intelligence engineering team(110+ members), I built the whole team from zero. The team includes backend developers, frontend developers, App developers and QA. Our team is supporting all the products of Shopee as well as some other products of the SEA group. The features include

• Promotion Item Search
• Recommendation(including item, voucher, banner, brands…)
• User Tag Intelligence
• Computer Vision
• Listing Intelligence
• Machine Translation
• Audio intelligence
• Identify verification(KYC)

and so on. We are supporting business lines including Promotion, User, Marketing, Order, Listing, Seller… We are improving the core indicators of the e-commerce platform and the iteration efficiency of both business projects and technical projects.

We also developed our in-house AI infrastructure from zero, such as:

• vector retrieval engine (based on Faiss, ScaNN, and redis Search)
• inference framework(based on Nvidia triton)
• feature KV middleware(kvRocks)
• data ingestion middleware

We also built our own AI API gateway to manage all our team’s AI services, including ACL, tenant quota management, sync & async protocol support, and 3rd party AI services(like ChatGPT, Google Translation…) management for the company.

Before this, I was leading the common shared service team to build middlewares and frameworks for the whole company. We built the common middlewares for the company including

• go service framework (it is the only backend service framework that supports the core marketplace systems including user, item, order, messaging…)
• smq (a golang message queue library, has 1000+ users in the company)
• gorm & gdbc (an in-house golang ORM library and a golang database connection pool library)
• ha-memcached (a high availability memcached client library)
• ssp (high performance and 10B records level query middleware)
• proxy server (long-connection tcp proxy server between shopee users and shopee backend services)
• … All these middlewares are initiated by my team and released & iterated with amazing efficiency, performance and stability. It was widely used in the company and successfully supported the Shopee business growth from 2019 to 2024. Our solution is one of the major contributors improving Shopee service SLA as well as supporting the big promotion events like 11.11, 12.12, 9.9, and so on.

Highlights

• 2023 Performance: A-(Partial Outstanding)
• 2022 Performance: A+(Outstanding)
• 2021 Performance: A+(Outstanding)

I was in the business architecture team under the IES group of Bytedance. We built core backend services and solutions for several products including

1. TikTok
2. Douyin
3. Huoshan
4. Live Stream
5. FaceU

My team was focused on the efficiency direction and responsible for environment management, traffic management, and tracing solutions for commonly used middleware by business teams. I was leading the environment management team and released non-live environments(BOE) and pre-releasing environments(like PPE) to support traffic control and business verification under different environments.

Highlights

• Release Environment and traffic governance for IES products

I was responsible for the DMP platform design, development, and maintenance for the e-commerce group in 58. I led the whole team and finished the 1st feature mining, store, analysis, and online retrieval services in 58. The platform was widely used in different components of our e-commerce platforms like Ads retrieval (targeting), recall, ranking, and creativities. I was also the team lead of the membership(VIP) system, promotion system, and enterprise management system. We also built the beta version machine learning platform with 2 teammates.

Highlights

• 3rd prize in 58 Programming Contest, 2018
• Feature Platform(ETL)

I was responsible for the SCM platform development and maintenance for TMall Supermarket, and Lirengou(俪人购). I worked on several components including purchase, item, inventory, order, return & refund. And also the core developer of the settlement system of the entire SCM platform.

Highlights

• DoD on 11.11 promotion day(2015)

I was one of the core developers to develop the 1st 3d action mobile game in Netease. We used our self-developed 3d game engine to develop the core action systems and also used scaleform(flash) as the game UI engine. And later we also introduced Cocos2d-X to support the 2d components and UI features. I finished several sub-systems of the game including account, package(bag), sprite(pet), equipment, skill, and so on. I also developed several auxiliary tools to help our teammates do real-time updates for the properties of the monsters and players in the game, auto-importing UI resources generated by Cocos2d-X framework solutions, and so on. I gained a lot of experience in UI engines, resource optimization, and animation effects.

Highlights

• Action Mobile Game

I was responsible for the development and maintenance work for QQ Mail. I collaborated with my teammates and developed several new features including batch operating(Select, Delete…), Star Mail, Mail Tags, Important Mail, Mail Calendar & Events, Ads email aggregation, and so on. I also developed the email thumbnail cache service and it was used in the email preview. I also developed several features for the early versions of WeChat like “Add to favorites”.

Highlights

• QQMail & WeChat
• Drifting bottle