Bingran You — Memory Palace
This is the text index of the 3D memory palace at /palace. The same content is also available at the regular site pages linked below. The 3D scene is rendered with WebGL; this text layer exists for search engines, AI agents, JavaScript-free browsers, and assistive tech.
About
I am Bingran You (Chinese: 尤炳然), a PhD candidate in Applied Science & Technology at UC Berkeley, advised in the Haeffner Lab.
I build reliable AI systems — agent infrastructure, evaluation harnesses, and applied AI products that need to behave under noisy real-world conditions.
I run trapped-ion experiments in atomic, molecular and optical physics — integrated photonics for individual ion addressing, ion-photon interfaces, and 3D-printed micro ion traps for scalable hardware.
Both tracks share one craft: turning complex, noisy systems into something that behaves on purpose.
Focus
Agentic Builder
- Agent skills and tool use, with an emphasis on evaluation that mirrors real workflows.
- Productivity agents that triage notifications, dispatch background work, and stay out of the way.
- Open-source benchmarks for measuring agent capability and safety in simulated workspaces.
Ion Trapper
- Adjoint-optimized integrated photonic circuits for individual trapped-ion addressing.
- Temporally multiplexed ion-photon interfaces via fast ion-chain transport.
- 3D-printed micro ion trap technology for scalable atomic-physics platforms.
- Trapped-ion Ramsey interferometry probing fundamental physics of single-ion vibrational modes.
Pages
Projects — Agentic Builder
- SkillsBench — A benchmark for evaluating how well AI agents use skills.
- first-tree — A Git-native context layer for decisions, ownership, and shared team knowledge.
- DoWhiz — Agent-native product for getting work done across email, chat, documents, and related tools.
- DeepTutor — An AI research assistant built on Zotero for cited answers, figure and formula understanding, and multi-paper comparison.
- mews — Local GitHub notification daemon that triages your inbox and dispatches Codex or Claude Code work for allow-listed repos while you sleep.
- smolclaw — Seeded mock environments for testing agent behavior in realistic workflows.
- SBTI CLI — An offline CLI for testing agent behavior with bundled logic and exportable results.
Projects — Ion Trapper
- bem — Scientific computing code for boundary element and fast multipole methods in Python.
- artiq_photonics_integration — An ARTIQ control framework for photonics integration experiments.
Papers — Agentic Builder
Papers — Ion Trapper
Education
- University of California, Berkeley — PhD Candidate in Applied Science & Technology (2022 — Present, Berkeley, California). Graduate research in atomic, molecular and optical physics — trapped ions, integrated photonics, and ion-photon interfaces. · Haeffner Lab
- University of Chinese Academy of Sciences — BS in Physics, Minor in Computer Science (2018 — 2022, Beijing, China). Undergraduate training in physics with a strong computational foundation across mathematics, programming, and scientific problem solving. · GPA 3.95 / 4.00 · Rank 1 / 54
Blog
- Zero-Human Company and One-Person Company: Two Names for the Same Shift (2026-05-08) — These phrases are near-synonyms. Both name the same 2024–2026 organisational pattern: a single founder running a complete business with an AI labour stack of agents, SaaS, and APIs. About 90% of the meaning overlaps. The difference is which side of the same idea each phrase emphasises — and that choice is mostly a cultural tell about who is talking.
- Reading the Agent-Organisation Stack Through Paperclip (2026-05-08) — Paperclip's framing — agents as employees, the control plane as the company — is the cleanest way to see what an actually-running zero-human / one-person company needs. Here is how I read the stack today, the specific primitives the control plane has to provide, and how Paperclip relates to the Codex App and Symphony patterns most founders meet first.
- What I See Working — and Not — in the Zero-Employee Operating Mode (2026-05-08) — A case-by-case read of the canonical 2024–2026 references — Project Vend, Andon Market, Paperclip, ClawBank, Coinbase x402, Coinbase one-person teams, Medvi — with what each one actually shows and what it does not. Anchored in public projects rather than personal claims.
- A Project I've Been Turning Over: Trust Infrastructure for AI in Expert Work (2026-05-06) — Notes from brainstorming a project idea — what it would take to make AI advice trustworthy enough for real expert workflows, and the design rules I keep coming back to: grounding, abstention, approval-gated action, and the day-14 reopen test.
- A Curated, Vetted Skills Catalog (2026-05-04) — Why I open-sourced the skills my agents use — every one personally verified, scanned for safety, and shown alongside its license, author, and source. Plus how to send me a skill you think belongs on the list.
- SEO and GEO for a Personal Site (2026-05-03) — What I changed on bingranyou.com so it gets found by Google, Bing, and LLM answer engines — without changing how the site looks.
- Welcome (2026-05-01) — Why this site exists, and what to expect here.
Recent posts
Contact