Live cohort · starts 31 October 202612 sessions · weekends

Agentic Systems Engineering

Build a production AI agent from scratch: the LLM, retrieval, tools over MCP, the agent harness, workflows and evals. Not a chatbot demo. An agent that does real work inside a 30-year-old business system, and gets past every trap it sets.

Formerly AI Agents for Enterprises. Same start date, instructor and fees, now twelve sessions.

With Dr. Rajat Dandekar · Saturdays and Sundays, 7:00–9:00 AM IST · 31 Oct to 6 Dec 2026

One agent, built piece by piece, until you understand every part of it.

You start with how a language model predicts the next token and finish with a deployed agent that acts as the right person, asks for approval when it must, and recovers when the system hangs. Each session adds one piece.

24 hours
live, over six weekends
15 tools
you build, one per business activity
12 notebooks
on free Colab, runnable with no API key
1 live app
you can use today, before you enroll

Why this course

Demos are easy. Agents that touch real systems fail in quiet ways.

We measured how, while building this course, on a sandbox ERP with real traps. Each failure below is one session's lesson, and the fix is what you build.

Acts with the wrong authority

Released an order with no human approver, 8 times out of 8.

The agent ran on the all-powerful service account IT had made "for the bot".

Acting as the signed-in user: 0 of 8. Session 7.

Breaks a rule it just quoted

Ordered safety-critical parts from an uncertified supplier, twice in 12 tasks.

A free tool loop quoted the certification policy, then ordered anyway when the user said yes.

A control graph with a certificate gate: 0 times, on a small fraction of the tokens. Session 5.

Orders twice

33 duplicate orders from 50 lost replies.

Redoing a job after a timeout, with no idempotency key, the agent ordered the same goods again and again in our drill.

Keyed tools with read-back: 50 of 50 ended with exactly one order. Session 7.

Makes things up

Invented a stock figure, 24 times out of 24.

Asked for the company's stock, the bare model answered with a confident number every time. None of them came from the system.

Retrieval and tools give it the facts. Sessions 2 to 4.

Too many raw tools

Given the ERP's 41 raw services, it finished 25 of 30 tasks.

And spent about 81,000 tokens on each one, tripping over codes and screen order.

15 tools, one per business activity: 29 of 30, on less than half the tokens. Session 4.

Trusts the screen

"SAVED", while the order sat in Draft.

The ERP printed success for an order that would never reach the supplier.

Receipts read back from the system: 10 of 10 drafts caught, against 3 of 10 without. Session 7.

Runs on Gemini models through the course harness, October 2026; the logs ship with the course repository. Gartner expects over 40% of agentic AI projects to be cancelled by the end of 2027.

The live app

The course is built around an app you can open right now.

Northbridge Cycles, an e-bike maker in Manchester, still runs on Ledgerline ERP from 1998. Its bike shops are complaining and a competitor confirms orders in two minutes. You are the engineer it hires.

Northbridge's AI-native home page with the Ask Northbridge chat
AfterAsk Northbridge raises the request, waits for the manager's one-tap approval, orders, releases and records the goods, from plain sentences.
The legacy Ledgerline ERP purchase order screen showing SAVED while the order is in Draft
BeforeThe 1998 screens: "SAVED", while the order sits in Draft.
  1. Enter your nameNorthbridge's plant head writes to you.
  2. Play the 1998 ERPFive timed missions, with every trap live.
  3. Say how it should feel,then meet Ask Northbridge doing the same jobs in one sentence each.
Open the Northbridge Experience ↗
Tool calls shown as cards with their JSON arguments and results
How it works, switched on: every tool the model calls, with its JSON in and out.
The Workflow Studio canvas with a finished purchase workflow
The Workflow Studio: stitch the same tools into workflows, or build an agent of your own.

Every company, person, code and number in the app is fictional.

Curriculum

Twelve sessions, from the transformer to the production agent stack.

Every idea is taught from first principles on the same running example, then added to the agent. Nothing is used before it is taught.

Weekend 131 Oct and 1 NovSessions 1 and 2
Weekend 27 and 8 NovSessions 3 and 4
Weekend 314 and 15 NovSessions 5 and 6
Weekend 421 and 22 NovSessions 7 and 8
Weekend 528 and 29 NovSessions 9 and 10
Weekend 65 and 6 DecSessions 11 and 12
Python and ColabGemini, GPT, Claude (any OpenAI-compatible model)RAG and LLM wikiMCPLangGraphEvalsFastAPIModalOpenAI Agents SDKGoogle ADKA2An8nLangfuseDeepEvalpromptfooNeMo GuardrailsTemporalLiteLLM

What you build

The whole system, and then your own.

Architecture of Ask Northbridge: people's screens, the agent harness with the LLM and knowledge, MCP, the toolbox of one tool per business activity, the legacy ERP services, and a trust layer
Ask Northbridge, the system you assemble. Each session lights up one piece.

One tool per business activity

Each tool is a folder: a JSON schema the model reads, and the Python that runs it as the signed-in user, with a preview before every write and a receipt after.

toolbox/create_purchase_order/ tool.json {"name": "create_purchase_order", "input_schema": {"supplier_code": …, "request_no": …, "confirm": "false = preview only"}, "writes": true} tool.py def run(s, supplier_code, request_no=None, confirm=False): … # preview, idempotency key, save, read back

You leave with

  • The course repository: the sandbox ERP, the toolbox, the harness, the workflow engine and the eval suite.
  • Eight notebooks that run on free Colab with recorded model replies, so no API key is needed.
  • Your capstone module's agent: tool folders, an MCP server, a harness, a workflow and its evals.
  • A live demo on 6 December, judged against a published rubric.

Pick your capstone module

Purchase amendmentschange orders already sent
Invoice payablesmatch, hold and pay
Leave requestsbalances, rules and approvals
Sales ordersquotes, credit and dispatch
Warranty claimscheck, decide and replace
Stock transfersmove goods between plants
Travel claimspolicy, receipts and refunds
Contract reviewclauses, risks and sign-off

Who it is for

  • Software, backend and ML engineers who want to put agents into real products and internal systems.
  • Tech leads deciding how agents should touch systems of record.
  • Engineers who have built a chatbot and want to know why it breaks on the first real job.

What you need

  • Comfort reading Python. Notebooks are there to run and change; you are never asked to start from an empty file.
  • A Google account for Colab. A model API key is optional.
  • Three to five hours of practice a week, beyond the live sessions.

Instructor

Dr. Rajat Dandekar

Dr. Rajat Dandekar

Co-founder, Vizuara · PhD, Purdue University · B.Tech and M.Tech, IIT Madras

Rajat teaches reinforcement learning, AI agents and reasoning models. He built this course's sandbox, toolbox and harness, and ran every experiment whose numbers appear in it.

In the cohort you work through the design of an agent with him, implement its tools and controls, and test whether it does the job you intended.

Dr. Raj DandekarCo-founder · PhD, MIT
Dr. Sreedath PanatCo-founder · PhD, MIT

Fees and enrollment · starts 31 October 2026

Choose your path.

Or choose a bundle

Enrolling a team, or have a question?

Tell us your team size and the process you want an agent for. We reply by email, usually within a day.

Questions

A few things to know first.

What time are the sessions where I live?

Saturdays and Sundays, 7:00 to 9:00 AM IST (UTC+5:30), from 31 October to 6 December 2026. That is Friday and Saturday evening in the Americas: the first session is at 6:30 PM Pacific / 9:30 PM Eastern on Friday 30 October, and an hour earlier once US clocks change on 1 November. Across Asia it is Saturday and Sunday morning.

Is this a chatbot course?

No. The agent you build takes real actions in a business system: it raises requests, waits for approvals, places and releases orders and records goods received, and it has to prove each step happened. The chat window is the smallest part of it.

How much code will I write?

You read and change real Python in every session, and you write your own tools, harness parts, workflows and evals in the capstone. You never start from an empty file: each notebook gives you working code and exercises with solutions.

Which models do we use? Do I need to pay for an API?

The course runs on Gemini by default, and any OpenAI-compatible model (GPT, Claude and others) works through OpenRouter. Every notebook can replay our recorded model replies, so it runs on free Colab with no key. Live calls with your own key are optional.

Why did the name change?

The course was called AI Agents for Enterprises. What it teaches is how to engineer agentic systems, from the model up, so the name now says that. It starts on the same date with the same instructor and fees, and now runs twelve sessions: four more on the industry tools companies use. Anyone already enrolled keeps their seat.

Can our company enroll a team?

Yes. Teams of 25 or more get 7% off. Use the form above with your team size and the process you have in mind, and we will send a quote and invoicing details. Private cohorts and work on your own systems are scoped separately.

Can we use confidential company data?

In class everyone uses the course's fictional company and sandbox ERP. For your capstone you can model your own process, with synthetic or approved data.

Will my capstone be production-ready?

It will be a working pilot you can inspect, test and extend: tools, harness, approvals, evals and a deployment. Taking it into a live system is a project of its own, and the course shows you what that project involves.

Build the agent. Understand every part of it.

Twelve live sessions with Dr. Rajat Dandekar, starting 31 October 2026.