Haodong L.

Key Skills

Software

Other

Top Subject Matter

FinTech AI Application (Credit Card Customer Operations)

Financial Document Risk Review (Due Diligence)

Customer Service Call Analysis

Top Data Types

Text

Document

Audio

Top Task Types

Text Generation

Entity (NER) Classification

Transcription

Freelancer Overview

Equipped with end-to-end capabilities in building high-quality datasets tailored for LLM fine-tuning and RAG, with hands-on experience across financial, customer service, and risk management domains. Designed a Text-to-SQL training data pipeline for a credit card operations platform, transforming 90+ dimensions of heterogeneous financial data into structured training sets through schema injection, automatic SQL repair, and safety guardrails — bridging the gap from prototype to production-ready data agents. Built a complete data processing workflow from scratch for a call center analytics project, encompassing speech-to-text transcription and six-dimensional semantic annotation (business classification, sentiment analysis, complaint risk grading, etc.), along with a label management system that standardized annotation guidelines for reproducibility at scale. In merchant risk review, automated the conversion of unstructured due diligence documents into structured review data via OCR + LLM, with multi-layer quality validation powered by RAG and rule engines. Core delivery competency centers on a closed loop spanning "data standard formulation → data cleaning & construction → annotation quality control → model performance validation." Able to independently author and review data standards, defining annotation specifications, taxonomy systems, and quality baselines. Proficient in leveraging prompt engineering, structured output enforcement (JSON Schema constraints), and few-shot example design to improve annotation consistency and dataset quality. Experienced in using Dify workflow orchestration to automate annotation task distribution and validation, significantly reducing manual effort. With practical expertise in vLLM private deployment and inference service optimization, capable of going beyond data delivery to directly participate in model fine-tuning evaluation — offering integrated "data + model" one-stop delivery assurance.

Labeling Experience

Customer Service Call Intelligent Analysis Platform Tech Lead

AudioTranscription

Headed the architecture and rollout of a speech transcription and analysis platform for customer service calls. Built multi-threaded batch audio transcription and enabled multi-dimensional labeling such as business tagging, summarization, sentiment scoring, and competitor identification. Designed a label management system to streamline configuration and data export/import processes for annotated calls. • Automated transcription labeling for high-volume audio data with concurrent processing. • Implemented multi-layer analysis agents for call tagging, pain point extraction, and sentiment recognition. • Designed custom workflows for risk grading and complaint threshold configuration. • Managed labeled datasets for T+1 call retrieval and daily report generation.

2026 - 2026

Merchant Risk Intelligent Review System Tech Lead

DocumentEntity (NER) Classification

Directed the construction of an intelligent review system utilizing OCR and LLMs for due diligence document analysis. Oversaw entity extraction, semantic validation, and public opinion risk detection pipelines. Employed explainable reporting agents for document reviews and supported dual-track decision making combining rule-based and LLM reasoning. • Structured document ingestion and extraction tasks for entities relevant to merchant risk review. • Labeled and classified due diligence materials by merging OCR and LLM-based entity mapping. • Developed semantic validation workflows to improve labeling accuracy. • Deployed Review-and-Generate (RAG) agents for audit and manual review support.

2025 - 2026

AI-Driven Credit Card Intelligent Operations Platform Tech Lead

TextText Generation

Led the design and implementation of a multi-agent platform for supervised and generative AI tasks in credit card customer operations. Coordinated agents handling data exploration, customer segmentation, and marketing content creation with workflow orchestration tools. Designed processes where agents generated and validated marketing strategies using hybrid rules and generative techniques. • Orchestrated prompt generation and output validation for customer-targeted messaging. • Developed quality evaluation methods to reduce AI hallucination during data-to-content workflows. • Processed and categorized 90+ dimensions of customer data for agent task routing. • Integrated agent-driven content labeling and feedback aggregation for continuous model improvement.

2025 - 2025

Education

Work History