Brent Wei Liu

Brent Wei Liu

San Francisco Bay Area
5K followers 500+ connections

About

Built production LLM search ranking at Pinterest — fine-tuned Llama-3-8B (qLoRA)…

Articles by Brent Wei

Activity

Join now to see all activity

Experience

  • Pinterest Graphic

    Pinterest

    Palo Alto, California, United States

  • -

    Menlo Park, California, United States

  • -

    Sunnyvale, California, United States

  • -

    Beijing City, China

  • -

    Beijing City, China

Education

Licenses & Certifications

Courses

  • Advanced Software Engineering

    -

  • Analysis of Algorithms

    -

  • Introduction to Algorithms

    -

  • Operating Systems

    -

  • Principles of Database Management Systems

    -

Projects

  • Log Analyzer

    Implement "Log Analyzer" system:
    1.Events extracting from textual logs;
    2.Events dashboard;
    3.Failure prediction: extracting features using time window technology, training predict model using machine learning algorithms ( SVM, kNN, Naive Bayes, Random Forest and Assembly ...), monitor system and predict failures.
    Technology: D3.js, JQuery, Bootstrap, Python, Django, Scikit-learn

    Other creators
  • Spatial Join

    Implement R-Tree Indexed Database and Spatial Join Algorithms.

  • B++ Tree Indexed Data Base

    Data Base include SQL Parser, Index Management, Record Management, Page Management, Buffer Management in it.

  • Operating System (NACHOS)

    -

    Implementation of Fork, Exec, Exit, Yield system calls and exception handling.
    Implemented and mastered concepts of virtual memory, swap files, thread synchronization and distributed mutual exclusion.

  • Mobile App&Game Search

    -

    Worked as the tech-lead of indexing team. Lead the following projects:
    a). App&Game Search Engine: Build the index of mobile software meta data and provide online search service.
    b). Authorization Mining: To improve the performance of app&game search engine ranking algorithm, applied learning to rank algorithm to generate a ranking score as authorization score for every mobile app&game. Then used this score in online search engine to re-rank the app&game instances. SVMrank was used in…

    Worked as the tech-lead of indexing team. Lead the following projects:
    a). App&Game Search Engine: Build the index of mobile software meta data and provide online search service.
    b). Authorization Mining: To improve the performance of app&game search engine ranking algorithm, applied learning to rank algorithm to generate a ranking score as authorization score for every mobile app&game. Then used this score in online search engine to re-rank the app&game instances. SVMrank was used in this project.
    Achievements: After the “Authorization Mining” project the search relevant of “Mobile App&Game Search” product has been increased from 81.5% to 90.2%.

  • Novel Vertical Search

    -

    Create a vertical search engine from beginning. Participated in following projects:
    a). Web Spider: Scrawl raw novel web pages from source sites.
    b). Structured Data Extractor: Parse raw novel pages, extract structural data and load the data into data base.
    c). Data Mining Platform: Manage and plug in data mining models to generated features of novels use in search engine rank system and recommending system. The data mining tasks like query count mining, click count mining, novel…

    Create a vertical search engine from beginning. Participated in following projects:
    a). Web Spider: Scrawl raw novel web pages from source sites.
    b). Structured Data Extractor: Parse raw novel pages, extract structural data and load the data into data base.
    c). Data Mining Platform: Manage and plug in data mining models to generated features of novels use in search engine rank system and recommending system. The data mining tasks like query count mining, click count mining, novel type aggregation, novel name aggregation … were complemented in the data mining platform.
    d).Novel Search Engine: Build the index of novels and provide online search service.
    e).Recommendation System: Include generalized recommendation system and personalized recommendation system.
    f).Key-Value Massive Data Storage System: Deployed a no-Sql data base named Mola to store novel content which cannot be indexed and stored by traditional data base system. Because the novel content data is huge and queried frequently. Those queries need quick response.
    Achievements: The PV per day of novel search has been increased to 100 million after 2 years team work, and the number of active user per day has been increased to 3 million.

Honors & Awards

  • Second Prize at China Undergraduate Mathematical Contest in Modeling

    -

Languages

  • Chinese

    Native or bilingual proficiency

  • English

    Full professional proficiency

More activity by Brent Wei

View Brent Wei’s full profile

  • See who you know in common
  • Get introduced
  • Contact Brent Wei directly
Join to view full profile

Other similar profiles

Explore top content on LinkedIn

Find curated posts and insights for relevant topics all in one place.

View top content

Add new skills with these courses