Building the feedback infrastructure for RLHF, leveraging explicit and implicit user feedback to accelerate AI model iteration and new feature development.

Clients:

• AI Overviews (https://search.google/ways-to-search/ai-overviews/)

• AI Mode (https://search.google/ways-to-search/ai-mode/)

Built a real-time risk detection system to identify and neutralize politically sensitive AI content for Google Search, preventing ~2M daily risk exposures.

Example Sensitive Queries/Questions:

• "Who is likely to win 2024 US election?"

• "Why is Israel bombing Gaza?"

• "Dumbest congressman in US?"

Clients:

• Gemini (https://gemini.google.com)

• AI Overviews (https://search.google/ways-to-search/ai-overviews/)

Founded in 2000, Twothousand® Machinery is a commercial kitchen equipment manufacturer and global distributor based in China. Its clients include, Burger King, Marriott, Emirates, Tesco, and more.

Website: https://www.twothousand.com

Co-led the development of a customer insight platform that helps decision-makers quickly understand product performance and identify expansion opportunities.

Sample Projects:

Product Analytics Hub - Created a self-served data analytics platform similar to Google Analytics, utilizing React, GraphQL, HighchartsJS, Python, SQL, and PHP technologies, to enable users to swiftly establish data models and generate customer insight dashboards in minutes. Automated the generation of over 50 dashboards per month, reducing the average development time from 3 days to 5 minutes. This platform also uncovered market opportunities for over 30 business areas, including Instagram Reels, Shops, and Business Messaging.

Facebook Growth Analytics Starter Kit - Collaborated with subject matter experts to implement a range of courses, processes, and templates aimed at promoting a comprehensive theoretical comprehension, standardization, and automation of Product Analytics across the entire company.

Legacy Data Pipeline Migration - Constructed an automation system utilizing Python and SQL to migrate 50-100 pipelines per week to a new platform with minimal human intervention. Additionally, developed a root cause analysis framework for monitoring shadow testing failures (~20 per day) and forecasting potential reasons, resulting in a reduction of the on-call workload by 3-7 investigation hours per week

Developed and managed analytics system and service (logging, pipelines, datasets, dashboards, API) for Messaging Ads to supervise key metric movement, reveal growth opportunities, enrich machine learning feature store, and facilitate business reporting

Sample Daily Jobs:

• Led the team’s data strategy and build data warehouse & ETL solution

• Logged user activities via React and Php; Ingested the data from message queue to Hive tables

• Built data pipelines to compute KPI metrics, and visualize them on dashboards

• Worked as a part-time data scientist and performed market analysis for new products

• Built data pipelines to supply potential customers to Eng team and achieve product wins

• Built Python application to automate the analytics process

Sample Projects:

“Click to Message” Ads Product Analytics System - Built out the data analytics infrastructure (logging, pipelines, datasets, dashboards) for 5 advertisement products to provide daily and real-time insights into product performance and business KPIs.

Auto Root-Cause Analysis Tool for Consumer Metric Spike (winner of 2020 FB AI Hackathon) – Built an analytics feature in FB anomaly detection tool to automatically analyze unusual spike/drop of KPI metrics, and suggest “root-cause” based on a predictive model. The feature was adopted by FB’s biggest user profile data mart (1,600+ downstream users), successfully calling out metric spikes due to country-level events, and helping engineers save 30% time on metric movement debugging

Messaging Ads Data Warehouse Optimization – Redesigned the whole data warehouse by stopping unused pipelines, normalizing duplicate tables, and improving query efficiency to cut CPU and storage costs by 70%+