Long-Horizon AI Task Agents & Reusable Agent Skills (SOPs)

📌 Project Overview

• Role: AI Agent Developer at Sihe Information Technology Co., Ltd.
• Focus: Autonomous LLM Agents, Multi-turn State Management, Agent Skills, Model Distillation.

⚙️ Key Technical Contributions

1. Long-Horizon Task Decomposition:
   • Engineered autonomous agent architectures capable of sustaining multi-step reasoning, dynamic replanning, and self-reflection across extended workflows.
   • Standardized reusable Agent Skills (Standard Operating Procedures) and tool interfaces.
2. LLM Mechanics & Model Distillation:
   • Investigated tokenizer behaviors (BPE, byte fallback) and domain-specific token representations.
   • Conducted Teacher–Student knowledge distillation experiments for lightweight, latency-critical inference.
3. Vibe Coding & Agent-Assisted Development:
   • Leveraged state-of-the-art coding agents (Codex, Hermes, pi) to achieve rapid prototyping, code refactoring, and test automation.