Peixing You manipulation · locomotion · navigation · dexterous hand · RL · LLMs · AI agent systems

CTO & Co-Founder — AuthenlyUSA

• Shipping full-stack features on AuthenlyUSA, Transnode’s AI credential & identity verification product for hiring and compliance workflows.
• Built admin review tooling: rejection notes, secondary staff review, document sorting/filters, and expiry-reminder flows used in live verification operations.
• Improved candidate subscription & billing UX — cancel-at-period-end, expiry reset, pricing/CTA consistency — across React/TypeScript surfaces on a production SaaS codebase.

Partners & programs

NSF I-CORPS™

IBM Startup Partner Plus

Notion for Startups

Atlassian for Startups

Apple Small Business Program

UC Launch Accelerator

Google for Startups Cloud

BBB Accredited Business

AuthenlyUSA trusted network

LinkedIn

indeed

glassdoor

ZipRecruiter

UC San Diego

National Student Clearinghouse

Parchment

CeCredential Trust

Checkr

Coursera

Regula

Member of Technical Staff

• Translated product needs into an LLM-powered multi-agent simulation system with Python & FastAPI; work supported acquisition of real paid startup customers.
• Integrated MinIO (S3-compatible) object storage for scalable multi-file user workflows across simulation sessions.
• Diagnosed long-context LLM blocking failures and implemented timeout/recovery logic, cutting worst-case pipeline blocking by 96.7% (300s → 10s).

Trusted by leading institutions

Harvard Kennedy School

MIT

BU Questrom

C10 Labs

NVIDIA Inception

Research — Dexterous Hand & Teleoperation

• Built a neural optimization pipeline (PyTorch + NVIDIA Warp) using forward kinematics and human finger-motion data to explore low-cost dexterous hand designs for manipulation and teleop.
• Closed the Sim2Real gap with SDF-based collision metrics; cut inter-finger collision cases in the physical 3D-printed prototype by 50%.
• Assembled a Gello-based teleoperation platform (YAM arm, Dynamixel actuators) and upstreamed CAD/spec fixes via open-source PR.

Organization

UC San Diego

Xiaolong Wang Lab

Hardware–Software Lead — AI Earphone Prototype (Lumen)

• Built a vision-enabled AI earphone prototype in three days: ESP32 camera, Bluetooth, speaker output, mobile triggering, and LLM connectivity.
• Led end-to-end hardware–software integration (ring wake-up, capture, BLE chunked transfer, phone-side processing) and refined product scenarios with investors and mentors.

Event

AdventureX 2025

LLM Agent Engineering / Testing Intern

• Built evaluation workflows and result-checking logic for an LLM spreadsheet-processing agent — reliability, format consistency, downstream compatibility.
• Scaled multi-threaded Python evaluation pipelines and structured test datasets to accelerate R&D decisions.

Organization

Baidu

Research Assistant — Perception & 3D Learning

• Designed an LLM-based active-perception policy with memory for indoor object localization (~3× efficiency vs. strongest baseline SPL).
• Co-developed SNAKE (NeurIPS 2022 Spotlight): shape-aware self-supervised 3D keypoint fields with up to 85% repeatability.
• Built IEEE CYBER 2024 Best Poster interactive grasping pipeline (ROS + RGB-D + point clouds) for crowded shelf retrieval.

SNAKE collaborators & venues

Tsinghua University

AIR, Tsinghua

Peking University

Intel Labs

MIT

Renmin University of China

NeurIPS 2022