Levi Lee

tensorOwl | Levi.

AI / LLM Engineer

Building RAG systems, agents, and the boring-but-real infrastructure behind them.

I'm a tensorOwl — half tensor, half owl, hatched from a TensorFlow typo I decided to keep. I build RAG systems, agents, and the unglamorous infrastructure that keeps them honest in production. A political-science grad turned self-taught engineer, now shipping LLM applications and co-founding a software startup in the space.

Writing

  1. 01

    Before You Reach for a Vector Database: Keyword Retrieval in the Agent Era

    Why do the strongest coding agents mostly reach for grep instead of a vector database? Notes on keyword retrieval, BM25, SQLite FTS5, Chinese tokenization, and why — in the agent era — retrieval is a workflow. Ending with the bigger shift from RAG to context engineering.

    Jun 1, 2026
  2. 02

    I vibe-coded this whole site from my phone

    From buying the domain on Spaceship and deploying on Cloudflare's free plan, to driving Claude Code from my phone, reviewing and merging PRs on GitHub, and wiring up GA4 and Google Search Console — I built this entire site without a laptop. How far phone-only "vibe coding" can actually go in 2026.

    May 30, 2026
  3. 03

    Python 101: The Syntax You Actually Need Before Your First AI Project

    Variables, conditionals and loops, lists and dicts, functions, and how to read a program — the core Python you'll actually use before building an AI project, with runnable examples. A follow-up to the NTU AI Club's first class.

    May 29, 2026

All writing

Selected work

  1. 01

    Social-Debate-AI

    Multi-agent LLM debate system on LangGraph, built to rigorously test — not assume — whether RAG, a GNN, and RL actually improve persuasion.

  2. 02

    Carbon Trading Experiment

    Real-time experimental-economics platform on oTree for carbon policy, with a live WebSocket permit market and an order-matching engine.

  3. 03

    Synthetic Difference-in-Differences

    Causal-inference library on PyPI implementing synthetic difference-in-differences with staggered adoption and event-study analysis.

All work

Background

Levi teaching a class at the NTU AI Club
Teaching at the NTU AI Club

Contact

Open to coffee chats