Skip to main content
LITT ACADEMY · AI IN PRACTICE · LIVE COHORT

Use AI in your practice — without betting your licence on it.

Six weeks, live, on the AI work a litigator actually does — research that survives a bench, judgment analysis, chronologies, case prep — and the verification habits that keep a citation that doesn't exist out of your written submissions.

See the curriculum
Your track
Live online·6 weeks·Built for Indian practice·No accreditation claimed
Why this course exists

Every lawyer has tried AI. Almost nobody has a method.

You already tried a chatbot on a research question. It gave you a confident paragraph and a case number. You had no fast way to know whether the case existed.

A general model · legal research question
Illustrative
> Authorities on whether a promoter's related-party pledge triggers disclosure
The position is settled. The Court held that a pledge in favour of a promoter group entity is a related-party transaction for disclosure purposes1, and the same reasoning has been applied to indirect holdings2. The contrary view has not been followed since3.
Not yet verified

The parties, citations and holdings below are invented for teaching. Any resemblance to a reported matter is coincidental.

1
Meridian Infrastructure Pvt. Ltd. v. Union of India(2021) 4 SCC 318
Verified

Reported, and it holds what the answer says it holds. This one you can cite.

2
Alcor Textiles Ltd. v. Commissioner of Income Tax(2019) 12 ITR 655
Superseded

A real matter. The provision it construes was amended after it was decided, so the ratio no longer reaches your facts.

3
Sunder Rao v. Kaveri Mills Pvt. Ltd.(2022) 8 SCC 774
Does not exist

Right court, right era, right shape of citation. It was never filed. The model produced the form of an authority without the substance.

In the answer above, all three read exactly the same. Nothing in the prose told you which one would survive. That is the entire problem. Everything else in this course is the habit that stops it reaching a bench.

The curriculum

6 weeks. 6 labs. Nothing you can't defend.

One live session a week, plus a lab you run between sessions. Every week is built on work you already recognise — the 6pm agreement, the Thursday hearing, the research note you can't vouch for — and every week ends with something you can use on Monday. Open any week below to see the outcome and what you walk away holding.

Week 1 of 6 · 90 min live + lab

How it actually works — and how it fails

Before any tool: what a language model is doing when it answers you, and why that machinery is perfectly capable of producing a citation that was never filed anywhere.

By the end of this week

You can look at any AI answer and say — before you've checked a single line — which parts are retrievable facts and which are guesses wearing the costume of a citation.

You leave withA one-page failure-mode card for your desk
The work this is about

You asked a chatbot for authorities on a limitation point. It gave you four cases in thirty seconds. You spent the next hour discovering that two of them don't exist.

A junior hands you a research note that reads beautifully. You have no way to tell which parts anyone checked.

You've quietly stopped using AI for anything that matters, because you can't tell when it's confidently wrong.

What we cover
What the model is actually doing when it answers — predicting the next word, not looking anything up
Why fabrication is a property of the machinery, not a bug you can prompt your way out of
The four failure modes: invented authority, superseded law, sycophancy (it agrees with your framing), and false confidence on genuinely open questions
Why the same question gives you a different answer on Tuesday — and what that tells you about relying on it
Retrieval-grounded versus open generation: the one distinction that predicts whether an answer is checkable at all
Where the fabricated-citation orders actually went wrong — the step the lawyer skipped, not the step the model took
The lab

Break a model on purpose. You'll have it invent a case, a section and a bench — and see how good the fake looks. Then run the same question through a grounded tool and watch what changes in what comes back.

What you'll be able to do

Six weeks from now, on your own litigation work.

01

Run a research question end-to-end and verify every authority you cite, in under a minute each.

02

Reconstruct a chronology from a mixed document set — emails, notices, agreements, scans — with every event tied back to its source.

03

Pull the ratio out of a long judgment, then check whether it has been followed, distinguished or overruled.

04

Keep a verification record you could hand to a judge who asks how you got there.

The lab built for your track
Week 3 — the citation you almost filed

You take a real research question, let a general model answer it, then verify each authority by hand. Whatever doesn't survive is the interesting part.

The core idea

The question isn't whether AI is good enough. It's whether you can check it.

Every task in your week sits somewhere on two axes: how badly it hurts if the answer is wrong, and how quickly you could tell. Where those two lines cross decides what you hand over — and what you never should. Here is that map for litigators.

Delegate
Delegate, then verify
Use with care
Do it yourself
Costly if wrongCheap if wrong
Hard to checkEasy to check

Hover a task to see where it lands. In week 5 you plot your own.

What you leave with

What you keep. None of it is a certificate.

Courses that end in a PDF change nothing on Monday. This one ends with documents you put to work.

Week 1
A failure-mode card

The one-pager that tells you, at a glance, which parts of any AI answer are checkable and which are guesses wearing a citation. The whole course in a page.

Week 2
Your AI-use policy

A written policy on what may be pasted into which tool, by whom, with whose sign-off. Drafted in week 2, stress-tested by week 6.

Week 3
A verification protocol

The check you run on every AI answer before it leaves your desk — fast enough that you'll actually use it, defensible enough to show a bench.

Week 4
A structured-brief template

The brief that gets you a usable first draft instead of a plausible one, written for the three documents you produce most.

Week 5
A delegation matrix + one workflow

Your own tasks plotted by stakes and verifiability — and one repeatable process for whatever eats the most of your week, timed to prove it was worth building.

Week 6
A supervision & rollout plan

How AI gets used across your team, who checks what, and what you'd show a court or a client who asks. The document you take to the partners.

Before you register

Don't take this course if —

We would rather lose the registration than the afternoon of somebody who wanted a different thing.

You want AI to practise law for you. It can't, and we won't teach you to pretend otherwise.

You want a list of prompts. Prompts age badly. The method here is verification, and it doesn't.

You want a vendor demo. LITT turns up in some exercises — so do the general models and the databases you already pay for.

You want a certificate more than a habit. There is no accreditation here and we claim none.

6 weeks
Live cohort, one session a week
90 min
Per live session, plus a lab
6
Deliverables — one every week
0
Client files you upload
One live session a week, plus a lab you run between sessions.·Sessions are recorded; cohort members keep access.·Cohort dates go to the register-interest list first.

Questions, answered.

Do I need to be technical?

No. There's no code and no maths. If you can read a contract, you can do this course. The difficult part is judgment, not software.

Is this really just training on LITT?

No. The course is tool-agnostic. LITT appears in some exercises because it's what we build and it makes the grounded-versus-generated distinction easy to show — but you'll also work with the general models and the legal databases you already subscribe to. The method is the point, and it transfers.

Is it accredited? Does it carry CPD credit?

No, and we make no accreditation claim. What you leave with is a policy, a protocol, a matrix and a working workflow — artifacts, not a certificate.

Do I bring my own client documents?

No — and please don't. Every exercise that touches a document runs on redacted or sample material. What you do bring is your own questions, your own tasks and your own week; a client's file never has to leave your office. Week 2 exists precisely so you understand why that rule isn't optional.

Are the four tracks four different courses?

No. One cohort, one curriculum, six weeks — the track decides which examples, labs and delegation map you're given, not what gets taught. You sit with the other tracks, and that's deliberate: the in-house counsel in the room is who the litigator is writing for.

Can my firm send a team?

Yes. Weeks 5 and 6 are built for it, where the work shifts from personal speed to standardising and supervising across a team.

What if I miss a session?

Sessions are recorded; cohort members keep access. The labs are where the learning actually happens, though — those are worth moving your calendar for.

The tools aren't the hard part. The judgment is.

Cohort dates go to the register-interest list first. Register on the litigators track and we'll come to you first when the next one opens.