Wenlong Dong

Designed and delivered an end-to-end Retrieval-Augmented Generation (RAG) system to enable semantic search and question answering across hundreds of enterprise SOP and policy documents. The solution improved compliance teams’ efficiency in retrieving accurate, context-aware information from large unstructured datasets.

KEY CONTRIBUTIONS
• Designed a full RAG pipeline: ingestion, preprocessing, chunking, embeddings, vector indexing, and large language model (LLM) response orchestration.
• Implemented scalable vector search using AWS OpenSearch Serverless for high-performance similarity retrieval.
• Integrated Azure OpenAI (GPT-4 class models) with prompt engineering to generate grounded, context-aware responses and reduce hallucination.
• Built a Streamlit-based UI for non-technical users to interact with the system in real time.
• Deployed solution on Amazon EC2 (Linux) and managed version control via Git.
• Collaborated with compliance stakeholders to translate business requirements into a production-ready AI solution.

Designed and delivered an end-to-end Retrieval-augmented Generation (RAG) system to enable semantic search and question answering across hundreds of enterprise SOP and policy documents. The solution improved compliance teams’ efficiency in retrieving accurate, context-aware information from large unstructured datasets.

KEY CONTRIBUTIONS
• Designed a full RAG pipeline: ingestion, preprocessing, chunking, embeddings, vector indexing, and large language model (LLM) response orchestration.
• Implemented scalable vector search using AWS OpenSearch Serverless for high-performance similarity retrieval.
• Integrated Azure OpenAI (GPT-4 class models) with prompt engineering to generate grounded, context-aware responses and reduce hallucination.
• Built a Streamlit-based UI for non-technical users to interact with the system in real time.
• Deployed solution on Amazon EC2 (Linux) and managed version control via Git
Collaborated with compliance stakeholders to translate business requirements into a production-ready AI solution.

A Redshift-based data model that consists of several tables and views of sales data built via dbt. The data objects are refreshed daily or monthly in Airflow. As the project designer and builder, I contributed to building dbt macros to export the tables and views to the S3 bucket as CSV files. We also created Anaplan CloudWorks jobs to consume the CSV files regularly.

Oversaw, as part of a team, the migration of Salesforce data from the source environment to the target environment. The client wished to separate part of its business into an independent Salesforce environment.

I set up the primary Python framework and built the initial version of the data extraction process—from Salesforce to Python DataFrame. I created the complete solution for duplicate records identification and merging dup records. I designed and developed the parallel computing process for comparing huge amounts of data as well as the grouping logic based on Graph theory. I also designed and built many SQL Server objects, including views, stored procedures, and functions.

I designed and completed a medical data validation platform with Excel VBA independently. I implemented complex validation rules within the Excel modules so that users could have data automatically and entirely validated in Excel.

This platform has been accepted and used for the data collection process worldwide.

A property data ETL solution project aimed at manipulating existing ETL data flow to fit new government requirements. I was one of the primary SQL Server and SSIS solution developers and completed approximately 50% of the development tasks.