Wen Yuan

Evaluating production-scale AI-generated software solutions across Python and TypeScript repositories to improve engineering reliability, architectural consistency, and implementation quality. Designed multi-step agentic prompt workflows and built evaluation methodologies for hallucination detection, reasoning quality, edge cases, and system-design consistency.

Performed LLM response quality and coding-agent reliability evaluation using prompt engineering, RLHF-style feedback, and rubric-based scoring. Identified failure modes such as hallucinated APIs, weak context usage, unsafe refactors, and flawed reasoning chains, and supported model alignment through structured evaluation datasets.

Built an enterprise RAG and semantic analytics platform with FastAPI, Airflow, Docling, OpenSearch, and LlamaIndex, processing 50K+ document chunks and records. Improved hybrid retrieval precision@5 by 35%+ versus keyword-only baselines and reduced manual refresh effort by 70%+ while keeping common top-k retrieval under 500ms.

Delivered an AI-enabled full-stack platform using TypeScript, Node.js, Express.js, MongoDB, JWT, Stripe, and PayPal, supporting 1K+ user/project records. Integrated LLM-powered semantic search and assistance features, reducing manual content lookup and repetitive support workflows by 40%+, and improved AI task completion reliability to 95%+.

Owned analytics automation for Creamistry China franchise operations by consolidating POS, delivery/order, inventory, and sales data into repeatable reporting workflows. Built Python/SQL ETL pipelines that reduced manual Excel-based reporting by 60–70% and created dashboards that reduced ad hoc reporting requests by 40%+.

Contributed to enterprise software systems by developing Java/Kotlin modules, backend integrations, REST API workflows, and authentication logic. Improved reliability and reduced repeated release-blocking issues by 25–30% by implementing request validation, service orchestration, edge-case handling, and cross-service debugging.

Built backend-focused web modules for a large online services marketplace, including listings, user profiles, order/workflow tracking, admin tooling, and SQL-backed reporting views. Improved marketplace reliability and performance by resolving production issues and optimizing backend routes/queries to improve response times for common workflows by 20%+.

Master of Science in Computer Science (2023–2026), partnering with clients on AI engineering work involving LLM evaluation, retrieval-augmented generation (RAG), semantic search, and reliability testing.

Bachelor of Business Administration in Marketing (2009–2012) from Chongqing University.