You're going to learn how to work with AIwith business
Connect.AI Skill Map.
Connect.AIA real-world example.
Everyone open up your
Outlook email.
InboxThree screens and you’re in.
Your UConn address, and the temporary password from that email.
Replace the one you were issued. Eight characters minimum.
Photo, name, contact details. This is your directory entry.
The Connect.AI Network.
Everyone in the program, with a direct line to each of them — email, phone, and a message button. Your own card is whatever you just filled in.
We built an MCP.
A standard connection between an AI assistant and a system — its data, its documents, what it can do. Ours is read-only.
ChatGPT
Where the lectures live.
Every session’s deck gets uploaded here as the semester runs — and the same material is accessible through the MCP.
Four movements. Open the one your session is in.
Every session in it, with a Ready badge once the deck is up.
Its own page, start to finish. Arrow keys, or swipe on a phone.
Put it on your home screen.
Open connect-ai.world in Safari, tap Share, then Add to Home Screen — it opens full-screen, like an app.
The square with the arrow, in Safari’s bottom bar.
Connect.AIconnect-ai.worldScroll the sheet if you don’t see it, then tap Add.
One tap to the decks, the Network and your messages.
Building and deploying AI applications.
LLM foundations
How a model tokenizes what you give it and generates what comes back — and therefore when you can count on it, and when it will fail.
Grounding models with data
A model only knows what is put in front of it. Grounding puts the business’s own facts there — the company Palantir built.
Building agentic systems
When one call is not enough. Everything that actually works is three layers — the loop on top, and what holds it up.
The agentic loop
What the model does on its own, between your ask and the answer.
Context & knowledge
What it holds while it works — and what it goes and looks up.
Connectors
How it reaches the real world — the tools it is actually allowed to call.
Evaluation-driven development
You cannot ship what you cannot measure — and this system will not give the same answer twice. Two loops, and one calibrates the other.
in the loop
Slow and expensive — and the thing every other eval gets calibrated against.
evaluations
Cheap enough to run on every single change — once the human loop has calibrated it.
Machine learning foundations
Enough of the underlying machinery to reason about a model instead of guessing at it.
Training vs inference
What was learned once, and what is being computed every request.
Data & features
Models inherit whatever is in the data — including its gaps.
Generalisation
Why something can be right on your examples and wrong in the field.
Metrics
What the numbers actually measure, and what they quietly leave out.
The story of Connect.AI.
- 01Eliyahu starts a business consulting for financial services firms.
- 02Conversations with UConn administrators — Dean Greg Riley — turn it into a real thing.
- 03The university accredits it.
- 04Connect.AI gets built.
- 0580 applications come in.
- 06We interview 40. We let in 20.
- 07A deal with the CTSBDC brings in two businesses — and we are not limited to theirs.
- 08Here we are today, doing the actual thing.
Engineering
Present, build, and ship your first site.
Town Hall
Session 07Present your plan, exchange ideas, and pressure-test your build with the cohort.
Build & Submit
Sessions 08–09Show progress, submit your live site, then start applying the skills to real business work.
Engagement
Scope the work. Own the delivery.
The Scope
Sessions 13–14Scope the work, frame the value, and use Plan mode to map the build before you commit.
The Delivery
Session 15Transition ownership — the accounts, the docs, and the training so the business runs it solo.
Meet the businesses.
The partner businesses join. Your group rotates in to embed with a real Connecticut business — and from here, everything you build is for them.
Connect.AI