Tao Ren / Tempest · AI Engineer & Researcher
Building agents for complex work.
I'm Tao Ren (Tempest), an AI engineer and independent researcher. I build agents for document-heavy workflows and software engineering, and study how language models edit and generate structured content.
At Newfront, I work on Nova Agent for insurance operations. Previously at TikTok, I built an LLM coding agent for Objective-C-to-Swift migration with custom evaluation harnesses. My research includes FineEdit, TreeDiff, and Infant Agent.
Building agents
Nova Agent · Newfront / WTW
An internal AI coworker for insurance operators. I work on document-heavy workflows, generated contracts, company-style email drafts, tool use, and human-in-the-loop review.
A task can involve more than 50 PDFs, policy and submission documents, generated contracts, and internal services. My work connects tools, document processing, structured outputs, and operator review with the constraints of an insurance workflow.
TikTok coding agent
Developer tooling · Code migration
Built an internal LLM agent for Objective-C to Swift migration, with generation loops and custom evaluation harnesses.
Selected research
FineEdit
Findings of EMNLP 2025
Research on precise, instruction-driven editing in structured domains including code, LaTeX, and database languages.
TreeDiff
SURGeLLM at ACL 2026
AST-guided code generation with diffusion language models, treating code as a structured object.
Infant Agent
Agent research · arXiv preprint, 2024
Tool-integrated agents with task-aware functions, memory retrieval, and cost-aware API usage for multi-step tasks.
Product experiments
I also build products to explore how people interact with AI: AgentShelf, a hackathon prototype for AI commerce readiness; Obi, a frontend for a general-agent product; and Live Subtitles, an app for streaming speech transcription and translation.
How the work connects
My engineering work asks how agents can complete useful tasks with tools, documents, and human review. My research asks how models can make precise changes and generate outputs that respect structure. Evaluation connects the two: the output has to hold up in the workflow where it will be used.
Earlier work at Ant Group, Ansys, and Pittsburgh gave me experience with production services, mobile applications, and research software. That engineering foundation shapes how I build AI systems today.
My research and engineering journey · Technical background
Writing
Let's connect
For research, agent engineering, and collaboration, email me or find me on GitHub and LinkedIn.