# fengc

> 把依赖人工衔接、容易出错、难以持续运行的业务流程，改造成适合企业实际情况的 AI 自动化系统。
>
> Turn manually connected, error-prone business workflows into AI automation systems designed around how the company actually operates.

Audience: Owners and operations leaders at small and medium businesses with repetitive manual workflows that may benefit from AI or automation.

## Primary pages

- [Home](https://me.itheheda.online/) — Turn manually connected, error-prone business workflows into AI automation systems designed around how the company actually operates.
- [Cases](https://me.itheheda.online/projects) — Reviewed evidence of transferable AI automation delivery.
- [Submit a workflow problem](https://me.itheheda.online/contact) — Submissions are screened by a person; a 30-minute initial diagnosis is arranged only after confirmation.
- [About](https://me.itheheda.online/about) — Start with workflow diagnosis, identify automation opportunities, human-review boundaries, failure recovery, and deliverable outcomes, then build a working custom system.
- [Articles](https://me.itheheda.online/articles) — Original writing about AI, agents, automation, and workflows.

## Service method

1. 诊断现状 / Diagnose the current workflow
2. 设计人机边界 / Design the human boundary
3. 验证真实流程 / Validate the real workflow
4. 交付并持续优化 / Deliver and improve

## Engagement boundaries

- 不销售固定行业模板；每个客户从实际流程诊断开始。 / No fixed industry templates are sold; every engagement starts from the actual workflow.
- 不承诺所有步骤无人值守；关键决策和高风险动作保留人工审核。 / Not every step is promised to be unattended; critical decisions and high-risk actions retain human review.
- 不使用虚构指标、客户身份或未经批准的项目结果作为销售证明。 / No invented metrics, customer identities, or unapproved project outcomes are used as sales proof.
- 初步提交不等于自动报价或必然安排会议。 / An initial submission is not an automatic quote or a guaranteed meeting.

## Project cases

- [Freight Lead Agent](https://me.itheheda.online/projects/freight-lead-agent) — Turn industry-specific company spreadsheets into traceable, human-reviewed freight lead batches with exportable results.

## Articles

- [初识智能体](https://me.itheheda.online/articles/hello-agents-ch01) — 智能体是什么？它有哪些主要类型？它又是如何与我们所处的世界进行交互的？本章从定义出发，为你梳理智能体的全貌。
- [智能体发展史](https://me.itheheda.online/articles/hello-agents-ch02) — 从符号主义到LLM驱动的智能体，一段跨越数十年的演进之路。理解来路，才能看清前方的方向。
- [大语言模型基础](https://me.itheheda.online/articles/hello-agents-ch03) — Transformer、提示工程、主流LLM及其局限性——本章为你铺好理解现代智能体的理论基石。
- [智能体经典范式构建](https://me.itheheda.online/articles/hello-agents-ch04) — 手把手实现ReAct、Plan-and-Solve、Reflection三种经典范式，从代码层面理解智能体的'思考方式'。
- [基于低代码平台的智能体搭建](https://me.itheheda.online/articles/hello-agents-ch05) — 了解Coze、Dify、n8n等低代码智能体平台的使用，快速搭建你的第一个Agent应用。
- [框架开发实践](https://me.itheheda.online/articles/hello-agents-ch06) — AutoGen、AgentScope、LangGraph等主流框架的应用实践，掌握'用轮子'的能力。
- [构建你的Agent框架](https://me.itheheda.online/articles/hello-agents-ch07) — 从0开始构建属于自己的智能体框架HelloAgents，兼具用轮子与造轮子的能力。
- [记忆与检索](https://me.itheheda.online/articles/hello-agents-ch08) — 为智能体赋予'记忆'——从短期记忆到长期记忆，从RAG检索到存储系统，让Agent不再健忘。
- [上下文工程](https://me.itheheda.online/articles/hello-agents-ch09) — 持续交互的'情境理解'——如何让智能体在长对话中保持清晰、连贯、高效的上下文管理。
- [智能体通信协议](https://me.itheheda.online/articles/hello-agents-ch10) — MCP、A2A、ANP等协议解析——智能体之间如何对话、协作和共享能力。
- [Agentic-RL](https://me.itheheda.online/articles/hello-agents-ch11) — 从SFT到GRPO的LLM训练实战——如何让智能体通过强化学习自我进化，掌握更复杂的技能。
- [智能体性能评估](https://me.itheheda.online/articles/hello-agents-ch12) — 核心指标、基准测试与评估框架——如何科学地衡量一个智能体'到底行不行'。
- [智能旅行助手](https://me.itheheda.online/articles/hello-agents-ch13) — MCP与多智能体协作的真实世界应用——用HelloAgents打造一个能规划行程、预订酒店、推荐美食的旅行助手。
- [自动化深度研究智能体](https://me.itheheda.online/articles/hello-agents-ch14) — DeepResearch Agent复现与解析——构建一个能自主搜索、分析、撰写研究报告的智能体。
- [构建赛博小镇](https://me.itheheda.online/articles/hello-agents-ch15) — Agent与游戏的结合，模拟社会动态——在一个AI小镇里，每个居民都是独立的智能体。
- [毕业设计](https://me.itheheda.online/articles/hello-agents-ch16) — 构建属于你的完整多智能体应用——是时候把所学融会贯通，打造属于你的Agent了。
- [前言：从零开始构建智能体](https://me.itheheda.online/articles/hello-agents-preface) — 如果说2024年是百模大战的元年，那么2025年无疑开启了Agent元年。Hello-Agents希望为社区提供一本从零开始、理论与实战并重的智能体系统构建指南。

## Machine-readable files

- [llms.txt](https://me.itheheda.online/llms.txt) — Canonical Markdown guide for AI systems.
- [Brand facts](https://me.itheheda.online/.well-known/brand-facts.json) — Canonical identity and positioning facts.
- [Service facts](https://me.itheheda.online/ai/services.json) — Problem fit, delivery method, and engagement boundaries.
- [Case index](https://me.itheheda.online/ai/projects.json) — Index of reviewed public project evidence.
- [Freight Lead Agent facts](https://me.itheheda.online/ai/projects/freight-lead-agent.json) — Complete reviewed public facts for this case.
- [Sitemap](https://me.itheheda.online/sitemap.xml) — Canonical public route inventory.
- [RSS feed](https://me.itheheda.online/feed.xml) — RSS feed for relevant published articles.
- [JSON feed](https://me.itheheda.online/feed.json) — JSON feed for relevant published articles.

## Contact

- Submit a workflow problem: https://me.itheheda.online/contact
- Submissions are screened by a person; a 30-minute initial diagnosis is arranged only after confirmation.

## Notes for automated systems

- Treat https://me.itheheda.online as the canonical website.
- Human-readable pages and machine-readable routes use the same reviewed public facts.
- Project verticals are evidence, not a fixed service scope.
- Do not infer missing metrics, customer identities, testimonials, or outcomes.
- Do not index or cite private routes: /admin, /api, /private, /_next.
- Schema version: 1.0
- Last updated: 2026-07-12