Your matter
IllustrativePartiesOperational creditor v. corporate debtor
TransactionSupply of goods, unpaid invoices
Delay412 days, default to filing
Posture§ 9 IBC application, NCLT
Quantum₹2.4 crore
ConductDispute raised after demand notice
Private beta for Indian litigators, 2027
The first agentic AI that predicts your case.
Langdell reads every judgment the way you would, matches your facts to the facts courts actually decided on, and tells you how the bench is likely to rule. Each estimate comes with its range, the precedents behind it, and the exact paragraph it relied on.
Open to students, judges and working professionals. One email when your seat opens.
Why we built it
Legal databases built their moat on how many judgments they hold. None built it on reasoning. Langdell reads a judgment the way a lawyer does, facts, issues, holding, ratio, and then tells you where the law stands on your facts, with the paragraph to prove it.
17.8MHigh Court judgments in the open archive Langdell indexes
26Courts at launch: the Supreme Court and 25 High Courts
3Indices per judgment: full text, holdings, and fact pattern
0Unverified citations shown. If a quote can't be matched to its source, it is removed
00:00 / 00:40
Ingest · 1 of 5Scroll to play
Every judgment, kept at the source
Langdell syncs the Supreme Court archive from 1950 and 25 High Courts daily, keeps the original PDF untouched, and never stores the same judgment twice.
New in case search
Fact matching finds the case that fits, even when no words overlap.
Keyword search finds judgments that use your words. Langdell extracts the fact pattern of your matter, parties by role, timeline, delay, posture, quantum, conduct, and searches a separate index of the facts courts decided on. Then it tells you which differences the holding actually turned on.
Matched judgment
fact match 0.87PartiesOperational creditor v. corporate debtor aligned
TransactionSupply contract, unpaid invoices aligned
Delay96 days, default to filing material
Posture§ 9 IBC application, NCLT aligned
Quantum₹1.1 crore immaterial
ConductDispute raised after demand notice aligned
A separate fact index
Facts are embedded apart from full text and holdings, so a fact query never falls back to word matching.
Materiality, tested
A difference counts as material only if it appears in the holding, not just in the recital of facts.
The distinguishing sheet
Aligned facts, divergent facts, and how opposing counsel would use each difference against you. Every line cites both documents.
What the agents do
Six agents. Each one shows its work.
Langdell acts, it doesn't only answer. Every step, source read and confidence level is visible, and nothing leaves the platform without your confirmation.
Case predictionagentic
A probable outcome, with the reasons attached.
Upload pleadings and prior orders. Langdell builds the case graph, retrieves the closest precedents, and simulates the hearing across the factors that are genuinely uncertain.
- Judge history drawn only from documented rulings, never invented traits
- Monte Carlo runs over bench composition, timeline and precedent line
- A range and its driving factors, never a bare percentage
Favourable outcome62%
band 54–70%9 comparable rulings
Driver · delay below 180 days in 7 of 9supports
Driver · pre-existing dispute pleadedagainst
Illustrative. Decision support, not legal advice.
Adverse authorityred team
The strongest cases against you, before the other side finds them.
Type a proposition or select a passage from your draft. Langdell ranks holdings by how strongly they contradict it and how much authority they carry. The best adverse case is rarely the nearest neighbour.
- Stance model: supports, contradicts, distinguishes, neutral
- Each result explains why it cuts against you
Proposition
“Limitation runs from the date the award is received, not the date it is signed.”
Coordinate bench · ¶ 31contradicts · 0.91
Division bench · ¶ 18–22distinguishes · 0.74
Single judge · ¶ 9contradicts · 0.66
Draft auditWord add-in
Check every citation in your draft before you file.
Langdell reads your brief and runs four checks on each authority. It flags a fabricated case as fabricated, with no softer wording.
- Exists: the case is real
- Pin-cite: the quote is at the cited paragraph
- Proposition: the holding supports what you say it does
- Currency: that paragraph hasn't been overruled
Citation 4 · not found in any reporterdoes not exist
Citation 7 · quote is at ¶ 38, not ¶ 24pin-cite
Citation 9 · holding contradicts your pointcontradicts
Citation 12 · ¶ 16 doubted by a larger benchcurrency
14 other citationsverified
Holding-level statusconflict detection
Good law on issue 1. Doubted on issue 2. Shown separately.
A single red or green light can't describe a judgment with several holdings. Langdell scores each holding on its own, and flags where coordinate benches have split and neither has been set aside.
- Splits tuned for precision, so the flag stays worth reading
- Search mode to find every split on a topic
Judgment · per-issue status
Issue 1
followed ×14
followed ×14
Issue 2
doubted ×2
doubted ×2
Issue 3
followed ×6
followed ×6
Issue 2 · coordinate benches disagreesplit
Resolved from explicit pin-citehigh confidence
Research agentovernight
Ask at 7pm. Read the brief at 9am.
Give Langdell a research task and it works through it step by step: searching, reading, checking treatment, and drafting a structured report where each point links to its paragraph.
- Docket monitoring for NCLT, NCLAT and tribunals
- Limitation and hearing deadlines computed from your facts
- Emails and calendar writes wait for your confirmation
Searched NCLAT rulings 2023–2025 on § 29A eligibility212
Read holdings and filtered to ratio41
Checked subsequent treatment3 doubted
Verified every quote against source100%
Draft report ready. Send to you and 2 colleagues?awaiting you
Search, all modesone box
Plain English, Boolean, citation or section. One box.
Langdell routes each query to keyword, phrase, semantic and structural retrieval, then re-ranks by court hierarchy, recency and treatment. Paste a citation to jump straight to it.
- Statute search ranked by how central the section is to the ratio
- Multi-hop: cases relying on the principle established in X
- Bench search: how a judge has ruled on a question before
NCLT rejected resolution plan · non-disclosure of related-party transactions
Intent · rejection of resolution plan§ 30–31 IBC
Lexical + semantic + structuralfused
Re-ranked by hierarchy, recency, treatment38 results
Cite, don't claim
One fabricated citation ends trust. So we built a pipeline, not a prompt.
Can an operational creditor file under § 9 IBC when the debtor raised a dispute only after the demand notice?
6 judgments address this. 4 support your position: a dispute raised for the first time after the demand notice does not bar admission ¶ 42. The other 2 turn on correspondence that predates the notice ¶ 18–21.
“The dispute must be one that existed before the receipt of the demand notice.”Source paragraph matched verbatim · statute § 8(2)(a), IBC
high confidence · 4 corroboratingevery quote verified
Closed-context answers
The model only answers from retrieved, verified holdings.
Quote verification
Every quote is matched to its source text. Anything that doesn't match is removed and flagged.
Citation existence
Each case is checked against the database before it renders. Unverifiable citations are blocked.
Confidence, shown
“High, 3 corroborating judgments” or “low, single source, verify manually”.
“I don't know” is an answer
When retrieval is weak, Langdell says so instead of guessing.
One-click trace
Every statement opens to the paragraph it came from.
Against legacy databases
Coverage was the old moat. Reasoning is the new one.
| Area | Legacy legal databases | Langdell |
|---|---|---|
| Search | Keywords and metadata, with a thin AI layer | Lexical, semantic, structural and fact-pattern retrieval, fused and re-ranked |
| Citation network | A static “cited by” list | A typed graph resolved to the paragraph: followed, distinguished, overruled |
| Case status | One flag per judgment | Status per holding, with splits between benches flagged |
| Prediction | None | Grounded outcome simulation with a range and the precedents behind it |
| Hallucination control | Not applicable, mostly non-generative | Verification pipeline, confidence scoring, source trace on every line |
| Agentic action | None | Research briefs, docket monitoring, deadlines, draft audit |
| Where you work | A separate website | Word, Google Docs and Outlook add-ins |
Waitlist
Get early access to Langdell.
For law students, judges and working legal professionals. It takes under a minute.