< 技术博客 />

Multi-Agent LLM Systems: Lessons from Belief-Dynamics

What happens when multiple LLMs debate? Surprising findings on false consensus and hallucination amplification.

Zero-Knowledge Proofs for AI: Current Research

Exploring how zero-knowledge proofs can enable verifiable AI inference while preserving model confidentiality.

Building a Distributed Object Storage System

Design decisions and implementation details of building an S3-compatible distributed storage system from scratch.

Knowledge Distillation: A Comprehensive Guide

An overview of knowledge distillation techniques from my work on FisherKD-Unified, covering 12+ methods.

Designing a Multi-Language Code Search Engine

Deep dive into building CodeGraph - combining AST parsing, semantic embeddings, and graph analysis for intelligent code search.

Building an LLM-Assisted Theorem Prover

How I built ProofCraft - a tool that combines GPT-4 and Claude with Lean 4 for automated theorem proving.

Four Main-Track Acceptances in One Year: What Actually Worked

ICML, ECML PKDD, ICICS and ProbML in a single cycle — and the two habits that mattered more than any individual idea.

From Workshop to Main Track: What Changed in NanoZK

The VerifAI @ ICLR version and the ICICS main-track version answer the same question. Only one of them proved it at a size anyone cares about.

I Audited My Own Website and Found DEBUG=True in Production

Six findings from running a real security audit against this site. The worst one had been live for months and was caused by a single misspelled environment variable.