Kerrie W.

Key Skills

Software

Other

Top Subject Matter

Financial RAG / enterprise customer service (foreign exchange, payments, compliance)

Banking finance knowledge base for multilingual customer inquiry answering

Financial RAG knowledge base (bank distributed retrieval)

Top Data Types

Text

Document

Image

Top Task Types

Question Answering

Data Collection

Text Generation

Function Calling

Freelancer Overview

Large Model Application Development Engineer — Jijia Tech (evaluation dataset + LLM/RAG optimization). Brings 3+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include Internal, Proprietary Tooling, and Chroma. Education includes Bachelor of International Economics and Trade, North China University of Technology (2019). AI-training focus includes data types such as Text, Document, and Computer Code and labeling workflows including Evaluation, Rating, and Question Answering.

Labeling Experience

Dify Engineer — PetroChina Talent Evaluation Agent

OtherTextFunction Calling

Built an AI-powered talent evaluation agent using a workflow platform for enterprise evaluation scenarios. Prepared RAG configuration and prompts grounded in MySQL schemas to enable structured question answering and evaluation assistance. Developed and deployed agent workflows to support automated evaluation processes for talent management use cases.• Wrote environment setup documentation and designed RAG configuration. • Designed prompts based on MySQL schema structures for evaluation tasks. • Built workflows and deployed intelligent agents for enterprise evaluation scenarios. • Integrated RAG-driven responses within Dify-based agent execution.

2026 - Present

Custom AI Music Generation Agent (personal project)

OtherText Generation

Developed an AI music generation agent that enhances an open-source music model with a user-facing interface for creating editable MIDI outputs. Implemented chord and instrument configuration workflows and generated structured MIDI files for DAW use. Although not traditional annotation, the work involves preparing and generating training-like inputs/outputs through custom workflows for AIGC music creation.• Built a Gradio frontend for interactive music creation. • Used pretty-midi and NumPy to configure chords and instrument parameters. • Generated editable MIDI files to support customized chord progressions. • Delivered custom AIGC workflows enabling reliable generation of structured outputs.

2025 - 2026

Project — Bank Distributed RAG System (Milvus + Haystack)

DocumentData Collection

Architected and built a distributed bank RAG system using a large-scale vector knowledge base for retrieving relevant text for customer questions. Implemented an end-to-end pipeline from document loading and chunking to embedding retrieval and generation. Added data isolation, compliance filtering, and security controls to meet financial regulatory requirements.• Designed a distributed knowledge base with Milvus indexing for millions of vectors. • Implemented document processing pipeline: loading → chunking → embedding → retrieval → generation. • Applied compliance filtering and security controls for regulated data. • Achieved 85%+ retrieval accuracy and reduced repetitive customer inquiries by 30%.

2024 - 2025

Large Model Application Development Engineer — Jijia Tech (RAG pipeline + QA optimization)

TextQuestion Answering

Led end-to-end RAG development for a bank’s foreign exchange intelligent customer service, transforming knowledge into retrievable text for LLM question answering. Implemented terminology-aware chunking and multilanguage embedding and reranking to ensure the system answers accurately and remains compliant. Built containerized deployment components and FastAPI services to productionize the QA experience for enterprise users.• Conducted data cleaning and terminology-aware chunking for financial documents. • Selected multilingual BGE-M3 embeddings and applied reranking to improve retrieval relevance. • Implemented compliance filtering and security controls in the knowledge pipeline. • Deployed LLM services with Docker and API gateways for consistent one-click environments.

2024 - 2025

Large Model Application Development Engineer — Jijia Tech (evaluation dataset + LLM/RAG optimization)

Text

Built and evaluated LLM outputs for enterprise finance scenarios, using a curated dataset to measure accuracy and compliance. Created a 300+ evaluation dataset covering payment processes, bank differences, remit failures, and compliance rules. Tuned prompts and retrieval settings to improve performance metrics from 60% to 82% in production workflows.• Defined evaluation coverage for payment, compliance, and failure-case categories. • Performed prompt engineering and retrieval tuning cycles to raise answer accuracy and compliance. • Used multilingual embeddings and reranking to improve retrieval quality for customer service. • Supported end-to-end LLM integration validation for bank foreign-exchange intelligent customer service.

2024 - 2025

Education

Work History