Lily Zhang
C-Guard: A Constitution-Grid Instrument for Data-Efficient RL Alignment
Post-trained safety guard model with RL · COLM 2026 ER
My Most-Used Agent Skills
Track the most frequently used skills, prune the useless ones · 2026
Reward Shaping from Failure Taxonomy for RLVR Terminal Environment
Multi-turn harbor terminal environment · 2026
LosslessBench: Evaluating Inference Acceleration Impact on Model Behaviors
Blogpost + lossless100 dataset · 2026
Skill Claw: Self-Improving Robot Agents
RL environments that train code policy · 2026
Dual-Model Voice Agent (Walkie-Talkie)
Front-end model talks with the user; back-end model handles tool calls, search, and generative UI · 2026
SofaGenius: Orchestrate Agentic Post-Training
Top 15/13k finalist in Anthropic Global Hackathon · 2026
SuperGeneral: Compositional Tool Environments for Long-Horizon Agents
Frontier models on tool use, composition, and creation · 2026
Frontend Slides: AI-Native Presentation Generation
25K+ GitHub stars · 2026
Is Speculative Decoding All We Need?
Poster, AI Engineer World's Fair · 2026
Eureka: Feature Engineering as Agentic Code Generation
Poster, NeurIPS 2025 VLM4RWD · 2025
Building toward Embodied AGI: Solving the Data Bottleneck
General Chair, RoboGen Workshop, IEEE IROS · 2025
Deep-Seek: Deep Research Agent
LLM-powered retrieval engine that compiles entity tables from internet-scale sources · 2024
Continuous Learning Loop: Online / Offline Perception Systems
Seminar, University of Delaware · 2023
From Lab to Road: Turning Self-Driving Research into Production Features
Keynote at IEEE IROS Workshop · 2023
Deep Learning-based Data-centric Solutions
Computer Vision Summit 2023
Diversify AI Training Data
NVIDIA GTC Conference · 2023
Data Misconceptions: Challenges and Solutions
AI Accelerator Summit 2023
DeepPIC: Deep Perceptual Image Clustering for Identifying Bias in Vision
CVPR Workshops 2022