Emmanuel Chukwu

I love music and novel thinking. Looking for like minded people

[Emmanuel Chukwu @Emagzimus]

I’m currently transitioning into AI Engineering, with a focus on business applications.

So far, I’ve completed the foundational + Python programming sections (1–20) of an AI Engineer bootcamp, covering:

[Sabine Hossenfelder @skdh]

If we want to continue progress in science, we must be prepared to acknowledge that the correct description of nature will eventually simply be too complicated for humans to understand. I do not think it is wise to make human understanding a requirement. AI is the way forward.

[陈大黄 @realchendahuang]

很多人觉得，做AI变现最牛逼的就是做一个软件，然后人家给你订阅费，每个月给你付钱，打美元。

但是离真金白银最近的路径，还是去帮传统企业的小老板们做业务的数字化和智能化，也就是咱们最近比较时髦的FDE。

整个中国有14亿人，真正用过CodeX、Claude Code的，我觉得不超过1%。

绝大多数中小企业的老板，天天听到AI、看到AI，但他们总觉得AI这东西离他们还八丈远。自己想干，又害怕投入巨量成本，也找不到真正的人才。这里面是有一种需求真空的。

各位X上的人中龙凤们，你们的AI知识已经溢出了，我们要开始变现了。如果面对这样的中小企业主，我们该怎么办？

第一步是帮他们搭建信息骨架，建一套企业知识库。

大量中小企业平时的真实状态其实非常混乱：产品资料散落在几十个人的电脑桌面，销售话术全靠老带新口头传授，老员工一离职资料全被带走。

这一步的核心是梳理清楚公司有几个部门、每个部门下设哪些具体模块，哪些是制度规范，哪些是业务表单。在飞书或者Obidian建立一套结构化的知识库框架，把零散的文件全部模块化归拢。

光是把他们企业这一套信息数字化、归档化、牛逼化，这里面的价值就可以收1万块钱了。

第二步是打通数据链路，让企业知识库能够随着业务展开和有机生长，形成一个数据闭环。

很多员工每天的任务就是在微信、Excel表单，还有各个业务操作系统里面来回倒腾数据、倒腾表格、倒腾文件。这是非常低效、非常可重复化的行为。

我们这一步的核心就是要用MCP协议，或者直接通过API接口，把企业知识库和这一套东西对接起来，中间通过AI智能体来打通。

比如说很多单位会用钉钉，钉钉就有对应的API接口。如果不用钉钉呢？那就是手动填写这些入职手册之类的。这玩意儿该怎么办？就得好好地把它做数字化改造。

比如说销售在外部跟客户沟通的时候，这些东西就要有一套客户关系管理系统，以及对应的话术要点和整套接待SOP，这套东西一定要把它们数字化，填入到企业知识库里，然后形成一套可以反馈、可以复盘的流程。如果这里能搞通，就能再收个3万到5万块钱。

第三步，要给每一个岗位梳理职责，并对对应的业务流程进行复盘和智能化重构。

可以逐个跟每一个员工谈话，或者观察他们平时工作的细节，把他们岗位每天要做的事情列成一份清单，再以“岗位 + 职责”的方式列成表单。这样做AI改造的时候就非常有的放矢，可以逐个攻破。

对于可以重复、可以机械化的劳动，我们就用AI工作流代替；对于需要人判断、没那么机械化、需要主观灵性的工作，我们就给员工提供对应的skills，辅助他们提高效率，减少人力。

更深层的交付在于企业业务资产的沉淀与优秀能力复制。以外贸或者销售型团队为例，将沟通渠道里的客户聊天记录与通话录音接入自动化转写，清洗并结构化入库。

系统不仅能把成单率最高的业务员的跟单思路、话术结构沉淀成团队的通用知识库，让普通员工直接调用并拉齐能力，还能把素材生成、投放分析与线索跟进整合到一起，让管理层随时掌握真实的业务大盘与卡点。

做企业服务的核心，是把抽象的技术概念拆解成老板听得懂、算得清账的具象动作。

传统企业愿意掏几万甚至几十万买单，本质上买的是确定性的流程梳理、人工耗时削减与企业核心资产留存。

把这套分层交付方案做扎实，就是一门非常稳定且高客单价的企业数字化改造生意。

[Emmanuel Chukwu @Emagzimus]

9 down, 1 to go. Capstone next. Almost at the finish line and I'm buzzing 🚀

#MCP #ModelContextProtocol #AgenticAI #AIAgents #IBM

[Emmanuel Chukwu @Emagzimus]

Real use: a shared company MCP server where the AI reaches the files it needs and nothing else. Biggest lesson: building agents is as much about trust and boundaries as capability.

[Emmanuel Chukwu @Emagzimus]

Labs 3 & 4: remote servers over streamable HTTP, filesystem roots as security boundaries, and sampling, where the server asks the model for help only after the user approves.

[Emmanuel Chukwu @Emagzimus]

Lab 2: built an MCP server with tools (write/delete files), resources (read files, list dirs) and prompts (code review, docs), plus logging, progress and user elicitation. Real use: an assistant that works in your project folder and asks before deleting anything.

[Emmanuel Chukwu @Emagzimus]

Lab 1: a LangGraph ReAct agent on GPT-5 talking to two MCP servers at once (library docs + Met Museum) over STDIO and HTTP. Real use: a dev assistant that pulls current docs instead of guessing from stale training data.

[Emmanuel Chukwu @Emagzimus]

MCP (Model Context Protocol) is a standard way for AI models to connect to tools, data and prompts. Build a server once, any compatible client can use it. No more custom glue code for every integration.

[Emmanuel Chukwu @Emagzimus]

Course 9 done: "Build AI Agents using MCP." That's 9 of 10 in the IBM RAG and Agentic AI cert. One course left, the capstone, and I'm properly excited. Here's what MCP taught me 🧵

[Emmanuel Chukwu @Emagzimus]

Biggest takeaway: there's no single best framework. The skill is picking the right one for the job, and sometimes combining them.

8 down, 2 to go 🚀

#AgenticAI #LangGraph #CrewAI #BeeAI #AutoGen

[Emmanuel Chukwu @Emagzimus]

Course 8 done. This one was a tour of the frameworks I'll actually be building with from here: LangGraph, CrewAI, BeeAI and AG2 (AutoGen). Here's what I took from each 🧵

[Emmanuel Chukwu @Emagzimus]

AG2 (AutoGen): built AutoMed, a team of agents simulating a medical consultation. One analyses symptoms, one suggests treatments, one pulls live medical data, one handles follow-up. Not a doctor, but real potential for triage in stretched clinics.

[Emmanuel Chukwu @Emagzimus]

The thread through all three: the leap isn't AI that answers. It's AI that checks itself, corrects itself, and knows when it can't answer.

7 down, 3 to go 🚀

#AgenticAI #LangGraph #LangChain #AIAgents #RAG

[Emmanuel Chukwu @Emagzimus]

Course 7 done: "Agentic AI with LangChain and LangGraph." Last course taught me agents can use tools. This one taught me they can think in loops. Built 3 projects 🧵

[Emmanuel Chukwu @Emagzimus]

That one felt the most real. Upload a 200-page report or a contract, ask a question, get an answer that's been checked rather than just generated. Legal review, dense analyst reports, research where a fake citation is a disaster.