AgenticMinds · LexStruct · Hacksplosion 2026 · Deloitte GenW.AI

Agentic Intelligence
for Indian Case Law.

LexStruct processes any Indian court judgment through 12 pipeline layers — extracting issues, arguments, precedents, applicable law, detecting argument clashes, verifying accuracy via Brave Search, calculating tax exposure, and generating a structured headnote in English and Hindi. Entirely on-premise via Deloitte GenW.AI.

Multi
Specialised Agents
12
Pipeline Layers
23
Innovations (I + N)
4
GenW.AI Modules
100%
On-Premise
The Problem

Legal research is broken. It does not have to be.

A single Indian court judgment can run 50–200 pages, written in dense legal language with zero structure. Lawyers read the whole document before analysis can begin. That is hours of expensive time — every single case.

Unstructured documents

Issues, arguments, precedents and verdict are buried in dense paragraphs — invisible unless you read every word.

Hours of manual reading

Lawyers spend hours getting through the document before any analysis begins. Expensive and unsustainable at scale.

Missed precedents

Without systematic search, similar past cases that strengthen or weaken an argument go entirely unnoticed.

No institutional memory

Insights from past cases stay locked in lawyers' heads, not in a searchable, growing database everyone can access.

LexStruct — Specialised agents on Deloitte GenW.AI

GenW.AI is Deloitte's enterprise agent platform. LexStruct deploys multiple specialised agents across four GenW modules — all on-premise inside your organisation's network. No data leaves your infrastructure.

Agent Builder — orchestration, extraction, verification, synthesis Realm AI — document ingestion, chunking, vector knowledge base, headnote Q&A App Maker — unified practitioner interface, case upload, review panel Playground — LexGraph visualisation, verdict trend charts, analytics dashboards

How It Works

From document upload to structured verdict

A multi-agent pipeline built on Deloitte GenW.AI — four modules working together to read, understand, and structure any Indian court judgment in under 2 minutes. Each agent has one specific job. All processing stays on-premise inside your organisation's network.

App Maker (UI Portal) Realm AI (Ingestion & KB) Agent Builder (Extraction) Agent Builder (Verification) Agent Builder (Synthesis) Playground (Analytics)
PHASE A — DOCUMENT INPUT & INGESTION 5 AGENTS — PARALLEL · SIMULTANEOUS PHASE C — VERIFICATION (AGENT BUILDER + BRAVE SEARCH) PHASE B — ANALYSIS (AGENT BUILDER) PHASE D-E — SYNTHESIS · ENRICHMENT · STORAGE · DELIVERY APP MAKER Upload Portal Drag-drop PDF / DOCX Up to 200 pages · Batch Practitioner initiates pipeline REALM AI Ingestion Agent Splits doc into 2K-token chunks 200-token overlap · builds KB Vector knowledge base created AGENT BUILDER Master Orchestrator Routes to 5 agents in parallel Manages pipeline flow state Coordinates all downstream work AGENT BUILDER — Issue Agent Identifies the legal question What type of case is this? GST / Income Tax / Property… Sarvam LLM · GPT-4 AGENT BUILDER — Petitioner Agent Extracts plaintiff arguments Reads only the petitioner's side Isolated context — no cross-bleed Sarvam LLM · GPT-4 AGENT BUILDER — Respondent Agent Extracts defence arguments Reads only the respondent's side Parallel to Petitioner Agent Sarvam LLM · GPT-4 AGENT BUILDER — Precedent Agent Maps cited past judgments Extracts all case citations Queries LexGraph N1 network LexGraph · Citation DB AGENT BUILDER — Judgment Agent Extracts court reasoning Why did the judge decide this? Preliminary verdict extracted Sarvam LLM · GPT-4 AGENT BUILDER + BRAVE SEARCH API Verification Layer Cross-checks every extracted field Brave Search API · scores 0–100% Score <80% → RAG rewrite & retry AGENT BUILDER — Good Law Checker Are cited cases still valid law? Walks LexGraph · flags overruled precedents Prevents arguing from dead cases in court . AGENT BUILDER — Argument Clash Detector Compares both sides head-to-head Labels each argument pair: Contradicted · Conceded · Novel GPT-4 comparison · Counter-arg RAG HITL if <80% AGENT BUILDER Condition + HITL Node All scores ≥ 80%? YES → proceed NO → lawyer reviews in App Maker Human approves before proceeding AGENT BUILDER Synthesis + Verdict Assembles structured case card Tags: Allowed / Dismissed Sarvam LLM · Hindi translation AGENT BUILDER Enrichment Agents Citation strength — PageRank Tax exposure in rupees IndicTrans vernacular agent AGENT BUILDER PostgreSQL Store All fields + confidence scores Indexed · searchable · auditable SQL Generation Agent · GenW DB PLAYGROUND Analytics Dashboard LexGraph citation network viz Verdict trend charts · Bench profiler Demand exposure · Outcome rates
Total pipeline time: under 2 minutes for any 50–200 page Indian court judgment  ·  100% on-premise inside Deloitte GenW.AI  ·  No data leaves your infrastructure

What each agent does — in plain terms

Every agent is built on Deloitte GenW.AI's Agent Builder module and runs inside the LexStruct pipeline. Here is what each one actually does and why it matters.

App Maker
Upload Portal
The practitioner-facing interface built in GenW App Maker. A lawyer drags and drops a court judgment PDF. One click starts the entire pipeline. No technical knowledge required.
Realm AI
Ingestion Agent
Realm AI reads the full document and breaks it into 2,000-token chunks with 200-token overlap. This builds the vector knowledge base that all downstream agents query. Without this, no other agent can understand the document.
Agent Builder
Master Orchestrator
The pipeline's central controller. Reads the Realm AI knowledge base and immediately routes work to all five specialist extraction agents simultaneously. Manages flow state and coordinates every downstream step.
Agent Builder
Issue Classification Agent
Answers the first question every lawyer asks — what is this case actually about? Identifies the exact legal category: GST, Income Tax, Property, Criminal, and assigns a confidence score so the pipeline knows how to process it.
Agent Builder
Petitioner + Respondent Agents
Two completely separate Agent Builder agents — one reads only what the plaintiff claims, the other reads only the defence. They run in isolated context so neither contaminates the other's legal analysis. This eliminates interpretation bleed.
Agent Builder
Precedent Agent
Extracts every case the court cited, then queries the LexGraph citation network to map relationships between judgments. Finds which past rulings were followed or distinguished. Eliminates days of manual precedent searching.
Agent Builder
Judgment Agent
Reads the court's actual reasoning — not just the outcome but the logical chain the judge followed to reach it. Extracts applicable statutes, sections, and the preliminary verdict. This is what lawyers need to cite in future arguments.
Agent Builder
Good Law Checker
Walks the LexGraph citation graph for each extracted precedent. Checks if any cited case has been overruled by a higher court, distinguished on facts, or superseded by legislation. Flags unsafe citations before the analysis is finalised.
Agent Builder
Argument Clash Detector
Compares petitioner and respondent outputs argument by argument using GPT-4. Labels each pair: Contradicted (direct clash), Conceded (one side didn't counter it), or Novel (new argument with no precedent). This is the core of legal strategy analysis.
Agent Builder + Brave Search
Verification Layer
Every single extracted field is scored 0–100% by cross-checking against live Brave Search API results. If any field scores below 80%, the system automatically rewrites using RAG and retries. This is what makes the output defensible in court.
Agent Builder + App Maker
Human-in-the-Loop (HITL)
If any field cannot reach 80% confidence after retries, the App Maker portal notifies a legal practitioner. The lawyer reviews and approves flagged outputs before the analysis proceeds. The AI never silently moves forward with uncertain data.
Agent Builder
Synthesis + Verdict Agent
Assembles all verified agent outputs into a single structured case card: legal issue, both sides' arguments, applicable statutes, precedents, reasoning chain, confidence scores, and a verdict tag — Allowed, Dismissed, or Partial.
Agent Builder
Enrichment Agents
Three parallel specialists: the Citation Strength Scorer uses PageRank to rank case authority, the Tax Demand Calculator queries the PostgreSQL store to compute rupee exposure range from similar past cases, and the Vernacular Agent translates via IndicTrans.
Agent Builder
SQL Generation + Storage Agent
Engineers INSERT queries to persist every extracted field along with its confidence score into the GenW.AI PostgreSQL store. Every processed case becomes part of the searchable institutional knowledge base — nothing is ever lost when a lawyer moves on.
Playground + Realm AI
Analytics Dashboard + Q&A
GenW Playground renders LexGraph citation networks, verdict trend charts, demand exposure visualisations, and bench-level profiling. Realm AI enables natural language Q&A directly against any case in the knowledge base.
Multi
Agent Pipeline
4
GenW.AI Modules
<2min
Per Judgment
12
Pipeline Layers
100%
On-Premise

Complete Agent Directory

Every agent — what it does, how it works

The full multi-agent LexStruct pipeline. Each agent has one job. All run on-premise inside the Deloitte GenW.AI network — no case data leaves your organisation.

#Agent NameGenW ModuleLLM / ToolRoleTag
01Realm AI IngestionRealm AI2K-token chunkerChunks PDF, builds Realm AI knowledge base
02Master OrchestratorAgent BuilderParallel routerRoutes chunks to 5 specialist agents in parallel, manages flow state
03Issue AgentAgent BuilderSarvam / GPT-4Extracts core legal question from judgment
04Petitioner AgentAgent BuilderSarvam / GPT-4Extracts petitioner arguments exclusively, isolated from respondent contextI7
05Respondent AgentAgent BuilderSarvam / GPT-4Extracts respondent arguments exclusively, runs parallel to Agent 04I7
06Precedent AgentAgent BuilderLexGraph N1Extracts all cited cases and applicable law sections
07Judgment AgentAgent BuilderSarvam / GPT-4Extracts court reasoning chain and preliminary verdict
08Argument Clash DetectorAgent BuilderGPT-4 comparisonLabels P vs R: Contradicted · Conceded · NovelN3
09Good Law CheckerAgent BuilderLexGraph Graph traversalWalks LexGraph, flags overruled / distinguished casesN9
10VL Query BuilderAgent BuilderGPT-4Converts extracted field to Brave Search queryI3
11Compare LLM AgentAgent BuilderGPT-4 Brave Search APIScores agent output vs web content 0–100%I1 I3
12HITL Condition NodeAgent BuilderCondition nodeFires HITL review only when any field score < 80%I2
13Synthesis + VerdictAgent BuilderSarvam / GPT-4Assembles headnote, tags verdict Allowed/Dismissed/PartialI8
14Citation Strength ScorerAgent BuilderLexGraph PageRankPageRank-style authority score per cited caseN2
15Counter-Arg GeneratorAgent BuilderRAG · PostgreSQLRAGs DB for top 3 adversarial responses per argumentN5
16Vernacular AgentAgent BuilderIndicTrans / SarvamTranslates headnote to Hindi / regional languageN10
17SQL Generation AgentAgent BuilderSQL gen PostgreSQLEngineers INSERT queries, stores all fields with confidence scoresI6
18Tax Demand CalculatorAgent BuilderPostgreSQLComputes rupee exposure range from similar case outcome patternsN7
19Section Lifecycle AgentAgent BuilderPostgreSQL · LexGraphYear-on-year interpretive drift per law sectionN4

Built with GenW App Maker

The AgenticMinds Platform — powered by LexStruct

Every screen your lawyers, analysts, and managers will use — built inside Deloitte GenW App Maker. The platform is named LexStruct in the product UI and connects directly to the full multi-agent pipeline. No technical knowledge needed to use any of it.



lexstruct.in · Dashboard
LexStruct
Good morning, Amaresh
Monday, 16 March 2026 · Orissa High Court
3 pending review
248
Cases
+12 this week
87%
Accuracy
+2.1% vs prior
1m 42s
Avg Time
Down 18s
94%
DB Match
Stable
Cases by Type — 8 Weeks
Recent Verdicts
Sharma v State
Property · 2h
Favourable
Rao v Rao
Family · 5h
Review
Singh Corp
Civil · Yesterday
Favourable
State v Mehta
Criminal
Pending
Outcomes
Pipeline Status
IngestionClassifyPrecedentArgsLawsVerdictOrchestrator
Dashboard
lexstruct.in · New Analysis
LexStruct
New Case Analysis
Upload any court document to begin processing
Drop your case file here
PDF, DOCX, JPG, PNG
Case type — Property / Criminal / Family…
Court name & jurisdiction
Year of judgment
Case Upload
lexstruct.in · Case Library
LexStruct
Case Library — 248 cases
Search by party, court, section…
Sharma vs State of Odisha
Orissa HC · 2024
Property
State vs Mehta — IPC 420
Supreme Court · 2023
Criminal
Rao vs Rao (Custody)
Family Court · 2024
Family
Singh Corp vs Union
Delhi HC · 2023
Civil
Case Library
lexstruct.in · Analytics
LexStruct
Most Cited IPC Sections
IPC 420
88
Sec 58 TPA
74
Art 300A
61
IPC 302
45
Outcome Distribution
Favourable 60%
Mixed 23%
Pending 17%
Analytics
lexstruct.in · Review Queue
LexStruct
Pending Review — 3 cases
Sharma vs State of Odisha
Case typeProperty Dispute
Petitioner strength8.2 / 10
Verdict confidence87% — Likely to succeed
DB match94% — Case #1
Add correction note
Lawyer Review
lexstruct.in · Admin
LexStruct
Team Access Control
AS
Amaresh Swain
Senior Lawyer
Full Access
PK
Priya Kumar
Associate
View + Upload
RM
Rohit Mishra
Analyst
Read Only
Audit Log — Today
09:14 · A.Swain uploaded Sharma vs State
09:16 · Precedent match: Case #1 (94%)
09:18 · Verdict generated — 87% confidence
10:02 · P.Kumar opened Case Library
Admin & Access Control
The platform includes an AI Assistant — interact with it live below
Step 1 — Case Details
Upload and Configure
Drop your case file here
PDF, DOCX, JPG, PNG — including scanned documents
document.pdf
Ready for analysis
Processing Pipeline

AgenticMinds Assistant

LexStruct Pipeline Ready — Multi-Agent

Built on
Deloitte GenW.AI
Welcome to AgenticMinds.

I am powered by LexStruct — a multi-agent pipeline running on Deloitte GenW.AI. Upload your case document on the left, fill in the details, and click Start Analysis.

All processing steps will run automatically and I will update you at each stage. You can also ask me anything right now.
AgenticMinds · LexStruct · Deloitte GenW.AI — All data stays within your organisation