Excelrs AI Agent Lab: six weeks, one deployed agent

The Excelrs AI Agent Lab is the current teaching row with a public week plan. Six weeks. Two evenings a week. Twelve live sessions. One deployed agent and an Excelrs certificate. Sandeep Jaiswal trains it. The listing price is $799 a seat; the first eight from that introduction are $499. Enquire first, then an invoice in USD. This post walks the plan that is already on that page. It does not change the price and it is not a recording of the sessions.

On 24 October 2026 the training row ai-agent-lab was still in the Ongoing bucket on /trainings (started 15 September, forty-two days). If you are reading this after that window, treat the page as the source for dates.

Who it is for

The lab is the same stack Excelrs already ran with campus batches, Gurukul, and LangGraph summers — live, in a US evening. It is a small batch, not a YouTube course. Recordings are backup. The product is the live room and the agent you ship.

You need enough Python or JavaScript to move, not a computer-science degree. If you have never put a file online, read From Prompt to Live Link first. The lab will not pause to teach you what a repo is.

Live sessions. Recordings are backup, not the product.

The six weeks, as published

Week 1 — Start where work actually is

How teams use AI day to day. Enough Python or JS to move. Cursor as the daily tool. First LLM call in your own repo.

Done looks like this: a repository that is yours, one successful model call, and a note of which model and which key path you used. Not a screenshot of someone else’s chat.

Week 2 — Prompts that hold

Chat versus one-shot. Structured output. A Q&A bot on your own notes.

This is where most beginners stay forever — a bot that talks. The lab pushes you to named fields and your own notes, so the answer can be checked.

Week 3 — RAG, your files, not the whole internet

Search plus citations. When not to use RAG. A bot that answers from your PDF or folder.

RAG means the model retrieves a chunk of your material before it speaks. It is the difference between “AI” and “AI that read the brief.” The week also says when not to use it. If the file is three paragraphs, paste the three paragraphs. A vector store is not a personality.

Week 4 — Tools, the model does a real action

Function calling. One API: calendar, sheets, or your app.

A tool is a function you wrote that the model is allowed to call. Calendar, sheets, or an endpoint from an app you own. Not twenty APIs. One.

Week 5 — Agents (LangGraph)

Multi-step work. Memory and failure. An agent that recovers from a bad answer.

LangGraph is how Excelrs has been teaching that loop on campus summers. An agent that cannot recover from a bad step is a demo. This week is about the recovery.

Week 6 — Ship and demo

Eval a few cases. Deploy a live URL. A five-minute demo plus the certificate.

If there is no URL, the lab is not finished. If you cannot break your own agent with two nasty cases, the demo is not ready.

Each week adds one ability. The certificate is for the URL at the end, not for sitting through week one.

What “one deployed agent” means

A finished lab project, in Excelrs language, has four parts:

1. A live link someone else can open.
2. A README that says the job in five lines.
3. Two or three eval cases — a normal question, a missing-file question, a question that should be refused.
4. A short demo you can do in five minutes without apologising for the setup.

It does not need a new company. It does not need a custom model. It does need to do one job on your files or your API.

How this differs from the other catalog rows

You do not need to have sat every older row. The lab assumes you will work, not that you collected every past certificate.

Campus internships in Jharkhand are a different door: campus program. Fees there are in rupees and the artifact is a university-facing certificate plus a profile. Do not mix those prices into this lab.

How to enquire without wasting a seat

The page is enquire first, then register and pay. Send who you are, your timezone, and one sentence about the agent you want to ship. If you cannot name the job, you are not ready for week 4.

Bring to week 1:

• A GitHub account and a repo you can push to.
• Cursor installed.
• A model key you are allowed to use.
• One folder of your own notes or one PDF you own.

Do not bring a client’s private data. Do not bring a prompt dump you do not understand.

The refused question is the one mentors remember. Teach the agent what it must not do.

Fees, as on the page

$799 per seat. First eight from the introduction: $499. USD invoice after enquiry. That is the listing. This post does not discount it, and it does not invent a second brand of certificate. The certificate named on the page is an Excelrs certificate for this lab.

Checklist

• You can recite the six week titles without opening the tab.
• You know recordings are backup.
• You can name the job your agent will do in one sentence.
• You have a repo, an editor, and a key before week 1.
• You know where to enquire: /ai-agent-lab and /trainings/ai-agent-lab.
• You did not paste a fee that is not on those pages.

If you want the campus version of this work — LangGraph summers, diploma tracks, a Dumka office floor — read the internship post next. If you want the protocol under week 4, read the MCP post. The lab is the one that asks you for a URL in week 6.