Skip to content
BranchesTags

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
.devcontainer
.devcontainer
 
 
.streamlit
.streamlit
 
 
core
core
 
 
data
data
 
 
pages
pages
 
 
utils
utils
 
 
.gitignore
.gitignore
 
 
README.md
README.md
 
 
app.py
app.py
 
 
requirements.txt
requirements.txt
 
 

Repository files navigation

⚖️ LitigatAI — AI-Powered Litigation & Dispute Resolution Platform

LitigatAI is a Streamlit-based AI platform built for Indian advocates, litigation interns, and legal researchers. It provides 10 integrated tools that automate the most time-consuming parts of legal practice — from extracting dates out of case files to drafting counter-arguments against an opponent's pleadings.

Table of Contents

  1. Project Overview
  2. Architecture
  3. The 10 Tools — Detailed
  4. Data Sources — What Is and Isn't in This Project
  5. Legal Coverage — Honest Assessment
  6. AI Models & API Stack
  7. Setup & Installation
  8. Configuration (.env Reference)
  9. Project Structure
  10. Limitations & Known Gaps
  11. Dependencies

1. Project Overview

Attribute Detail
Platform Streamlit multi-page web app
Primary LLM Google Gemini 2.5 Flash
Fast/Routing LLM Groq Llama 3.3 70B
Fallback LLM OpenAI API (gpt-oss-120b-onprem)
Case Law Source Indian Kanoon API (live)
Court Status Source eCourts (ecourts.gov.in) via public API
Hardcoded Legal Data 10 Indian Acts, ~60 key sections (see §5)
Document Formats Input: PDF, DOCX, images. Output: DOCX, CSV, JSON
Language Support Hindi, Bengali, Marathi, Tamil, Telugu, Kannada, Malayalam, Gujarati, Punjabi, Urdu, Odia → English OCR + translation

2. Architecture

┌─────────────────────────────────────────────────────────────┐
│                  LITIGATAI  (app.py)                        │
│                  Streamlit Web Application                  │
│                                                             │
│  Sidebar navigation → 10 tool pages                        │
│  Dashboard: tool cards + upcoming hearings widget           │
└──────────────────────┬──────────────────────────────────────┘
                       │
         ┌─────────────┼─────────────┐
         │             │             │
    ┌────▼──────┐ ┌────▼──────┐ ┌───▼───────┐
    │  pages/   │ │  core/    │ │  utils/   │
    │  (10 UIs) │ │ (logic)   │ │ (helpers) │
    └───────────┘ └─────┬─────┘ └───────────┘
                        │
              ┌─────────▼──────────┐
              │      llm.py        │
              │  ┌──────────────┐  │
              │  │ Gemini 2.5F  │  │  ← primary (text + vision)
              │  │ Groq Llama3.3│  │  ← routing / fast classification
              │  │ OpenAI GPT   │  │  ← fallback on Gemini 429
              │  └──────────────┘  │
              └──────────┬─────────┘
                         │
          ┌──────────────┼───────────────┐
          │              │               │
   ┌──────▼──────┐  ┌────▼────┐  ┌──────▼──────┐
   │ data/ (JSON)│  │Live APIs│  │ File I/O    │
   │             │  │         │  │             │
   │ bare_acts   │  │ Kanoon  │  │ pdf_utils   │
   │ court_cases │  │ eCourts │  │ docx_utils  │
   └─────────────┘  └─────────┘  └─────────────┘

Request flow for a typical tool:

  1. User uploads a file or types a query in a Streamlit page
  2. Page calls the relevant core/ module
  3. Core module extracts text (via utils/pdf_utils.py) and/or calls the Kanoon API (via utils/kanoon.py)
  4. core/llm.py dispatches to Gemini (primary) → OpenAI (fallback on quota exhaustion)
  5. Response is rendered in the UI and optionally downloaded as DOCX/CSV

3. The 10 Tools — Detailed

📆 Tool 1 — List of Dates

File: pages/1_List_of_Dates.py | Core: core/list_of_dates.py

Extracts every material date and event from uploaded legal documents and produces a court-ready chronological table.

  • Input: PDF, DOCX (text-based or scanned), plain text
  • OCR: Scanned PDFs → rendered page-by-page to PNG → Gemini Vision OCR. Image-based DOCX → embedded images extracted → Gemini Vision OCR.
  • Output: JSON array [{date, event, source, significance}], downloadable as DOCX table or CSV
  • LLM Task: Gemini extracts dates, describes event in legal language (≤25 words), identifies source paragraph, assigns legal significance
  • Prompt strategy: System prompt instructs model to be exhaustive — no date skipped. Approximate dates ("sometime in January 2022") are included with a note.

🔍 Tool 2 — Legal Research Assistant

File: pages/2_Legal_Research.py | Core: core/legal_research.py

Multi-turn chat interface for Indian legal research — ask follow-up questions and refine answers without restarting the session. Upload a judgment PDF/DOCX to find the exact paragraph that supports or contradicts your argument.

  • Chat interface: Uses st.chat_input — submits on Enter (no Ctrl+Enter), works natively on mobile. Full conversation history rendered with st.chat_message; follow-ups like "focus more on section 73" work naturally.
  • Step 1 (Groq): Intent classification — categorises query as STATUTE, CASE_LAW, PROCEDURE, or GENERAL. Extracts jurisdiction, relevant acts, and builds an optimised short-form Kanoon search query.
  • Step 2 (Kanoon API): Live search returns up to 10 relevant judgments with citations, bench, and publication date.
  • Step 3 (Gemini): Synthesis — structures a full answer with inline citations to Kanoon results. Falls back to AI legal knowledge for general principles where Kanoon results are tangential.
  • Document upload + paragraph Q&A: Upload any PDF, DOCX, or TXT judgment. Ask "which paragraph supports my argument about damages?" — Gemini returns exact paragraph numbers labelled SUPPORTS / CONTRADICTS / CONTEXT. Prior Q&A on the same document is passed as context for coherent multi-turn follow-ups.
  • File caching: Uploaded file bytes are stored in st.session_state so asking multiple questions about the same document does not require re-uploading.
  • Integration with Tool 5: The Prompt Enhancer's "Send to Legal Research" button navigates here and shows the enhanced prompt in a copy-able block above the chat.
  • Kanoon results displayed: Title, citation, bench, headline snippet
  • Sidebar controls: Max results slider (1–10), Clear conversation button, Prompt Enhancement expander

✅ Tool 3 — Authority Check

File: pages/3_Authority_Check.py | Core: core/authority_check.py

Verifies whether a cited Indian judgment is still authoritative — or has been overruled, distinguished, or limited.

  • Input: Citation string (e.g., Maneka Gandhi v Union of India, AIR 1978 SC 597) or case name
  • Step 1: Strips citation codes (AIR/SCC/SCR) to extract party names; searches Kanoon for the target case
  • Step 2: Uses word-overlap scoring (_best_matching_tid) to find the correct TID among search results — avoids picking an unrelated first result
  • Step 3: Searches citedby:{tid} to retrieve all cases that have cited this judgment
  • Step 4 (Gemini): Reads citing cases and classifies treatment:
    • GOOD LAW — consistently followed, no adverse treatment
    • CAUTIONED — followed with reservations or modified
    • DISTINGUISHED — distinguished on facts in some cases
    • OVERRULED — expressly overruled by a larger bench or higher court
    • UNKNOWN — insufficient data
  • Output: Verdict badge + full reasoning with specific citing case names

🌐 Tool 4 — Translation + OCR + Ask AI

File: pages/4_Translation_OCR.py | Core: core/doc_translator.py

Handles legal documents in any Indian language — extracts text, translates to English, and lets you chat with the document.

  • Supported input languages: Hindi, Bengali, Marathi, Tamil, Telugu, Kannada, Malayalam, Gujarati, Punjabi, Urdu, Odia, English
  • OCR pipeline:
    • Digital PDF → PyMuPDF (fitz)
    • Scanned PDF → page-by-page PNG → Gemini Vision OCR (up to 8 pages)
    • Image-based DOCX → extract embedded images → Gemini Vision OCR
    • Direct image (JPG/PNG) → Gemini Vision OCR
  • Language detection: Gemini identifies source language from first 500 characters
  • Translation: Gemini translates with legal terminology preserved, party names transliterated, section numbers exact. Adds Translator's Note for ambiguous terms.
  • Ask AI: Chat interface backed by Gemini — answers questions strictly from document content with inline paragraph/section references

💡 Tool 5 — Prompt Enhancer

File: pages/5_Prompt_Enhancer.py | Core: core/prompt_enhancer.py

Transforms a rough lawyer's query into a structured, high-quality LLM prompt for better research outcomes.

  • Input: Raw legal question (e.g., "can consumer file complaint against builder for delay")
  • Classification (Groq): Legal area assigned from: CONSTITUTIONAL, CONTRACT, CRIMINAL, CIVIL_PROCEDURE, CONSUMER, PROPERTY, FAMILY, ARBITRATION, LABOUR, TAX, IP, CORPORATE, BANKING, GENERAL
  • Output: Structured prompt with: Legal Area, Jurisdiction (state/HC/SC), Relevant Acts, Specific Question, Key Concepts, Desired Output Format
  • Integration: "Send to Legal Research" button navigates to Tool 2 and displays the enhanced prompt in a copy-able code block at the top of the chat — paste it into the chat input to research immediately

📋 Tool 6 — Document Drafter

File: pages/6_Document_Drafter.py | Core: core/doc_drafter.py

Drafts Indian legal documents with auto-generated inline citations in professional court/arbitration format.

Supported document types:

Document When Used
Suit (Plaint) Civil suit to initiate proceedings in a Civil Court
Statement of Claims Arbitration claims under the A&C Act 1996
Written Statement Defendant's reply to plaint (Order VIII CPC)
Statement of Defense Respondent's reply in arbitration
Rejoinder Claimant/Plaintiff's reply to Written Statement / SoD
Vakalatnama Advocate's power of attorney to appear
Counter-Claim Defendant's independent claim (Order VIII Rule 6A CPC)
Application / Misc. Petition Interlocutory application within a pending suit

Professional formatting (updated):

  • Arbitration documents (Statement of Claims, SoD, Rejoinder): decimal paragraph numbering 1.1, 1.2, 2.1, 2.2 under Roman-numeral section headings (I. INTRODUCTION, II. PRELIMINARY SUBMISSIONS, A. / B. sub-headings)
  • Court filings (Plaint, Written Statement, etc.): plain Arabic numbered paragraphs 1., 2., 3. — each beginning with That…
  • Defined terms: First-reference definitions — "ABC Ltd. (hereinafter 'the Claimant')" — used consistently
  • Annexure cross-references: [Annexure A], [Annexure C-5] in body; ANNEXURES list at the end
  • Inline citations: [CITE: Act s.XX], [CITE: Case Name (Year) Citation], [CITE: verify — description] — extracted into a separate panel after drafting
  • Professional closing block: Filed through: [Counsel], Date, Place
  • Download: As formatted DOCX via utils/docx_utils.py

🏛️ Tool 7 — Court Date & Case Status Tracker

File: pages/7_Court_Tracker.py | Core: core/court_tracker.py

Manages your entire case docket in one place.

  • Local storage: Cases stored in data/court_cases.json (persists between sessions)
  • eCourts integration: Enter a CNR (Case Number Record) → fetches live status from services.ecourts.gov.in/ecourtindia_v6/cases/case_status_cnr. Uses plain buttons — no keyboard-submit shortcut (Ctrl+Enter) in this tab.
  • Fields tracked: Case name, case number, CNR, court, client, case type, status, hearing dates, notes, filing date
  • Dashboard: Homepage widget shows all hearings within 7 days with colour-coded urgency (red ≤2 days, orange ≤5, green otherwise)
  • Manual fallback: If eCourts API is unavailable, all data can be entered manually

📑 Tool 8 — Judgment Summariser

File: pages/8_Judgment_Summariser.py | Core: core/judgment_summariser.py

Converts a full judgment (any length, any format) into a structured legal brief.

  • OCR pipeline: Same as Tool 4 — handles digital PDFs, scanned PDFs (up to 10 pages via Gemini Vision), image-based DOCX
  • Structured output:
    • Case Details: Name, citation (AIR/SCC/SCR format), court, bench, date
    • Facts: All material facts in chronological order
    • Issues: Each question of law/fact framed by the court
    • Petitioner Arguments: Key contentions with statutory/case basis
    • Respondent Arguments: Key contentions with statutory/case basis
    • Held: Final order/decree precisely stated
    • Ratio Decidendi: Legal principle(s) established
    • Significance: Why this judgment matters for Indian law
    • Key Statutes Cited: List
    • Key Cases Cited: List
  • Fallback: If Gemini JSON fails, falls back to plain-text structured summary
  • Download: As DOCX brief

📖 Tool 9 — Bare Acts & Amendments Tracker

File: pages/9_Bare_Acts.py | Core: core/bare_acts.py

Browse and search Indian legislation with amendment history.

  • Source: data/bare_acts_data.json (hardcoded — see §5 for exact coverage)
  • Features:
    • Browse all 10 acts → select any section → view full text
    • View amendment history per section (which act amended it, when, what changed)
    • Full-text search across all sections and all acts simultaneously
    • Ask AI: type a question about any provision → Gemini answers with reference to the section text
  • Acts included: See §5 below for the exact list with section counts

🛡️ Tool 10 — Counter Arguments Drafter

File: pages/10_Counter_Arguments.py | Core: core/counter_args.py

A 4-step wizard that reads the opponent's pleading and drafts a complete Written Statement.

  • Step 1 — Upload: Upload opponent's plaint/statement of claims (PDF/DOCX)
  • Step 2 — Extract Claims (Gemini JSON): Identifies each legal claim/allegation as {claim_number, claim_text, legal_basis, remedy_sought}
  • Step 3 — Draft Counters (Gemini per claim): For each extracted claim, generates a counter-argument with: legal basis to deny, distinguishing cases, applicable exceptions, suggested factual rebuttals
  • Step 4 — Assemble (Gemini): Compiles all counter-arguments into a complete Written Statement in Indian court format with preliminary objections, para-wise reply, and prayer
  • Interactive: You can ask follow-up questions about any specific claim before assembling
  • Download: Final Written Statement as DOCX

4. Data Sources — What Is and Isn't in This Project

This is the most important section. The project uses three distinct types of data sources, each with different reliability and coverage guarantees.

A. Hardcoded JSON Data (data/bare_acts_data.json)

This file is a manually curated snapshot of selected provisions from 10 Indian Acts. It is static — it does not auto-update when Parliament amends a law.

Exact contents:

Act Year Sections Included Notes
Consumer Protection Act 2019 9 (Definitions, Jurisdiction, Complaints, Product Liability, Unfair Contracts, etc.) Covers District/State/National Commission hierarchy
Code of Civil Procedure 1908 9 (O. 6 R.17 Amendment, O. 7 Plaint, O. 8 WS, O. 39 Injunctions, s.9, s.10, s.151, etc.) Key provisions for civil litigation
Arbitration & Conciliation Act 1996 6 (s.7, s.8, s.9, s.11, s.34, s.36) Arbitration agreements, court intervention, enforcement
Limitation Act 1963 6 (s.3, s.5, s.14, Arts. 54, 58, 113) Includes extension and condonation
Indian Contract Act 1872 6 (s.2, s.13, s.14, s.73, s.124, s.126) Offer, acceptance, breach, indemnity, guarantee
Specific Relief Act 1963 4 (s.10, s.14, s.38, s.41) Specific performance, perpetual/mandatory injunctions
Transfer of Property Act 1882 5 (s.5, s.54, s.58, s.105, s.111) Sale, mortgage, lease
Bharatiya Nyaya Sanhita 2023 5 (s.3 General Exceptions, s.100 Culpable Homicide, s.101 Murder, s.318 Cheating, s.351 Criminal Intimidation) Replaces IPC 1860
Bharatiya Nagarik Suraksha Sanhita 2023 3 (s.173 FIR, s.242 Charge, s.479 Bail) Replaces CrPC 1973
Bharatiya Sakshya Adhiniyam 2023 4 (s.57 Electronic Records, s.22 Admissions, s.113 Presumptions, s.63 Secondary Evidence) Replaces Indian Evidence Act 1872

Indian Contract Act — Special Contracts (additional module):

Act Component Sections
Special Contracts s.148 Bailment, s.182 Agency, s.10 Wagering (void)

Total: ~60 sections across 10 Acts + 1 Special Contracts module

B. Live API Data

Source Data Provided Authentication Notes
Indian Kanoon API (api.indiankanoon.org) Full-text search of Supreme Court, High Court, and Tribunal judgments; case documents; citation relationships (citedby:) Requires INDIAN_KANOON_API_TOKEN (register at api.indiankanoon.org) Without token → mock/placeholder results. With token → millions of real judgments
eCourts (services.ecourts.gov.in) Live case status, next date, court, bench — for any case with a CNR number No auth required (public endpoint) Can return HTTP 400 for invalid CNR; occasional downtime

C. AI Model Knowledge (Gemini / Groq / OpenAI)

These models have broad legal knowledge from their training data, which includes:

  • Indian statutes (IPC, ICA, CPC, Constitution, etc.)
  • Supreme Court landmark judgments
  • Legal principles and doctrines

Important caveat: AI model knowledge has a training cutoff date and should not be relied upon as authoritative for recent legislative amendments or judgment updates. Always verify AI-generated legal content against primary sources.

5. Legal Coverage — Honest Assessment

What This Project DOES Cover

Area Coverage
Consumer law Consumer Protection Act 2019 — key sections (hardcoded) + live Kanoon judgments
Civil procedure CPC 1908 — 9 key sections (hardcoded) + live Kanoon judgments
Arbitration Arbitration & Conciliation Act 1996 — 6 key sections (hardcoded)
Limitation Limitation Act 1963 — 6 key sections (hardcoded)
Contract law Indian Contract Act 1872 — 9 key sections (hardcoded)
Property law Transfer of Property Act 1882 — 5 key sections (hardcoded)
Criminal law Bharatiya Nyaya Sanhita 2023 — 5 key sections (hardcoded); full IPC/BNS via AI knowledge
Criminal procedure Bharatiya Nagarik Suraksha Sanhita 2023 — 3 key sections (hardcoded)
Evidence Bharatiya Sakshya Adhiniyam 2023 — 4 key sections (hardcoded)
Civil relief Specific Relief Act 1963 — 4 key sections (hardcoded)
Case law (general) Live via Indian Kanoon API — covers SC, HC, Tribunal judgments (millions of documents)
Document languages 10 Indian languages → English OCR + translation via Gemini Vision

What This Project DOES NOT Cover (Not in the JSON or any live feed)

Area Status
Constitution of India ❌ Not in hardcoded data. AI models have constitutional knowledge but it is not a verified, section-by-section database.
Income Tax Act 1961 ❌ Not present
Companies Act 2013 ❌ Not present
GST / Indirect Taxes ❌ Not present
Labour & Employment Laws (ID Act, POSH, etc.) ❌ Not present
Intellectual Property (TM Act, Patents, Copyright) ❌ Not present
Banking & Finance (SARFAESI, FEMA, RBI Act) ❌ Not present
Real Estate (RERA) ❌ Not present
Family & Personal Laws (Hindu Marriage, MMA, etc.) ❌ Not present
Environmental Laws ❌ Not present
Motor Vehicles Act ❌ Not present
RTI Act, Prevention of Corruption Act ❌ Not present
State legislation ❌ Not present
High Court Rules / Civil Court Rules ❌ Not present
Recent amendments (post-2021) ⚠️ Partially — BNS/BNSS/BSA 2023 are included; others may be outdated in the hardcoded JSON. AI model knowledge may cover some gaps.
Law Commission Reports ❌ Not present
Notifications / Circulars ❌ Not present

Amendment Currency

  • The hardcoded JSON was manually curated. It does not auto-update.
  • The three 2023 codes (BNS, BNSS, BSA) are included as they replaced IPC, CrPC, and Indian Evidence Act.
  • For the most authoritative up-to-date text of any provision, always cross-check with:

6. AI Models & API Stack

Primary LLM — Google Gemini 2.5 Flash

  • SDK: google-genai (new SDK, not deprecated google-generativeai)
  • Model ID: gemini-2.5-flash
  • Used for: All synthesis, drafting, translation, summarisation, JSON structured output
  • Vision capability: gemini_vision() — used for OCR on scanned PDFs and image-based DOCX
  • Rate limits (free tier): 20 requests/day. After exhaustion, automatically falls back to OpenAI API.
  • Quota handling: Parses the retry in Xs delay from 429 error messages; sleeps and retries up to 3 times; then falls back to OpenAI.

Fast/Routing LLM — Groq Llama 3.3 70B

  • SDK: groq
  • Model ID: llama-3.3-70b-versatile
  • Used for: Intent classification (Legal Research routing), legal area classification (Prompt Enhancer), Kanoon query building — tasks where speed matters more than depth
  • Rate limits: Groq free tier is generous (~14,400 tokens/minute). Unlikely to be exhausted in normal use.

Fallback LLM — OpenAI API

  • SDK: openai (OpenAI Python SDK, base_url overridden)
  • Base URL: https://api.openai.com/v1
  • Model ID: gpt-oss-120b-onprem
  • Used for: Automatic fallback when Gemini hits daily quota limit
  • Capability: Text generation only — no vision/OCR capability. OCR-dependent tools (scanned PDFs) will fail if Gemini quota is exhausted.
  • JSON output: Prompted to return plain JSON; markdown fences stripped via regex before parsing.

Indian Kanoon API

  • Base URL: https://api.indiankanoon.org
  • Registration: api.indiankanoon.org — token required
  • Endpoints used:
    • POST /search/ — full-text judgment search. Returns {docs, found, categories}. Each doc has: tid, title, headline, fragment, citation, bench, publishdate, numcites, numcitedby
    • POST /doc/{tid}/ — fetch full judgment HTML by TID
    • POST /search/ with formInput: "citedby:{tid}" — all cases that cited a given judgment
  • Coverage: Supreme Court, all High Courts, various Tribunals (NGT, SAT, NCLT, etc.), District Courts (partial)
  • Without token: The platform falls back to placeholder results. Tools 2 and 3 still function via Gemini AI knowledge but without verified citations.

eCourts API

  • Base URL: https://services.ecourts.gov.in/ecourtindia_v6/
  • Endpoint: POST /cases/case_status_cnr
  • Auth: None (public endpoint)
  • Input: CNR number (18-character unique case identifier assigned by eCourts)
  • Returns: Case title, case number, filing date, stage, next hearing date, court, bench
  • Limitation: Only works for cases listed on the eCourts platform. Not all courts are integrated.

7. Setup & Installation

Prerequisites

Installation

# 1. Clone / navigate to the project
cd law-project

# 2. Create a virtual environment (recommended)
python -m venv .venv
source .venv/bin/activate        # Linux/Mac
.venv\Scripts\activate           # Windows

# 3. Install dependencies
pip install -r requirements.txt

# 4. Configure environment variables
cp .env.example .env             # or edit .env directly
# Fill in your API keys (see §8 below)

# 5. Run the app
streamlit run app.py

The app opens at http://localhost:8501 by default.

First Run Checklist

  • GOOGLE_API_KEY set — required for all AI features
  • GROQ_API_KEY set — required for routing/classification
  • INDIAN_KANOON_API_TOKEN set — required for live case law in Tools 2 & 3
  • OPEN_AI_API_KEY set — optional, enables fallback when Gemini quota exhausted
  • Upload a PDF or DOCX to test Tool 1 (List of Dates)

8. Configuration (.env Reference)

# ──────────────────────────────────────────
# REQUIRED
# ──────────────────────────────────────────

# Google Gemini 2.5 Flash (primary LLM + OCR vision)
# Get from: https://aistudio.google.com/app/apikey
GOOGLE_API_KEY=your_google_api_key_here

# Groq Llama 3.3 70B (fast routing/classification)
# Get from: https://console.groq.com
GROQ_API_KEY=your_groq_api_key_here

# ──────────────────────────────────────────
# OPTIONAL — but strongly recommended
# ──────────────────────────────────────────

# Indian Kanoon API (live case law for Tools 2 & 3)
# Register at: https://api.indiankanoon.org
# Without this, Tools 2 & 3 use only AI knowledge — no verified citations
INDIAN_KANOON_API_TOKEN=your_kanoon_token_here

# OpenAI fallback (when Gemini daily quota exhausted)
OPEN_AI_API_KEY=your_open_ai_key_here
OPEN_AI_BASE_URL=https://api.openai.com/v1
OPEN_AI_MODEL=gpt-oss-120b-onprem

# ──────────────────────────────────────────
# MODEL OVERRIDES (optional)
# ──────────────────────────────────────────

GEMINI_MODEL=gemini-2.5-flash
GROQ_MODEL=llama-3.3-70b-versatile

9. Project Structure

law-project/
│
├── app.py                        # Main Streamlit dashboard + hearings widget
├── requirements.txt
├── .env                          # API keys (git-ignored)
│
├── .streamlit/
│   └── config.toml               # Theme (navy #1B4F72, serif font), max upload 50MB
│
├── pages/                        # Streamlit multi-page routing
│   ├── 1_List_of_Dates.py
│   ├── 2_Legal_Research.py
│   ├── 3_Authority_Check.py
│   ├── 4_Translation_OCR.py
│   ├── 5_Prompt_Enhancer.py
│   ├── 6_Document_Drafter.py
│   ├── 7_Court_Tracker.py
│   ├── 8_Judgment_Summariser.py
│   ├── 9_Bare_Acts.py
│   └── 10_Counter_Arguments.py
│
├── core/                         # Business logic (no UI imports)
│   ├── config.py                 # Env loading, path constants
│   ├── llm.py                    # Gemini / Groq / OpenAI clients + retry/fallback
│   ├── list_of_dates.py          # Tool 1 logic
│   ├── legal_research.py         # Tool 2 logic
│   ├── authority_check.py        # Tool 3 logic
│   ├── doc_translator.py         # Tool 4 logic
│   ├── prompt_enhancer.py        # Tool 5 logic
│   ├── doc_drafter.py            # Tool 6 logic
│   ├── court_tracker.py          # Tool 7 logic + eCourts API
│   ├── judgment_summariser.py    # Tool 8 logic
│   ├── bare_acts.py              # Tool 9 logic
│   └── counter_args.py           # Tool 10 logic
│
├── utils/
│   ├── pdf_utils.py              # PDF/DOCX text extraction + image extraction
│   ├── kanoon.py                 # Indian Kanoon API wrapper
│   └── docx_utils.py             # DOCX generation (court-formatted output)
│
├── data/
│   ├── bare_acts_data.json       # 10 Indian Acts, ~60 sections, amendment history
│   └── court_cases.json          # User's case docket (local JSON storage)
│
├── uploads/                      # Temp storage for uploaded files
└── output/                       # Generated DOCX/CSV downloads

10. Limitations & Known Gaps

Legal Data Limitations

Limitation Impact
Only 10 acts hardcoded (~60 sections each) Tool 9 Bare Acts can only browse these 10 acts. Questions about other laws rely on AI knowledge only.
Constitution not in hardcoded data Constitutional questions answered by AI from training knowledge — not a verified section-by-section database
Amendment history is manually curated No auto-sync with Parliament / India Code. May be outdated for acts amended after the JSON was last updated.
BNS/BNSS/BSA 2023 included but not complete Only key sections included; full acts are much larger
No state legislation State-specific rules, amendments, or state-level acts not covered

API / Quota Limitations

Limitation Impact
Gemini free tier: 20 requests/day Platform will exhaust quota quickly in a demo or heavy-use session. Set OPEN_AI_API_KEY for fallback.
OpenAI fallback: text only Vision/OCR (scanned PDFs, image DOCX) requires Gemini. If Gemini quota is exhausted, OCR-dependent tools will error.
Kanoon token required for live case law Without token, Tools 2 & 3 return no real judgments — only AI knowledge
eCourts API: public, no SLA Can go down or return 400/503. Only covers courts on the eCourts platform.
Indian Kanoon coverage Not every HC/District Court judgment is indexed. Tribunal coverage varies.

Technical Limitations

Limitation Impact
Gemini Vision OCR: max 5-10 pages per document Long scanned judgments (100+ pages) will only have first 5-10 pages OCR'd
Text truncation: max ~20,000 characters sent to LLM Very long documents get truncated before synthesis
court_cases.json is local No multi-user or cloud sync — case data is per-machine only
No authentication The Streamlit app has no login — do not host publicly with sensitive client data
No real-time Parliament feed Cannot detect when a new amendment is enacted

11. Dependencies

Package Version Purpose
streamlit ≥1.35.0 Web UI framework
openai ≥1.30.0 OpenAI fallback LLM
google-genai ≥0.8.0 Gemini 2.5 Flash (primary LLM + Vision OCR)
groq ≥0.9.0 Groq Llama 3.3 70B (fast routing)
python-dotenv ≥1.0.0 .env file loading
pymupdf ≥1.24.0 PDF text extraction and page rendering (fitz)
python-docx ≥1.1.0 DOCX reading and generation
pandas ≥2.2.0 DataFrame for tables / CSV export
Pillow ≥10.0.0 Image processing
requests ≥2.31.0 HTTP calls to eCourts API
python-dateutil ≥2.9.0 Robust date parsing
reportlab ≥4.0.0 PDF generation (available, not currently used in active tools)
pydantic ≥2.7.0 Data validation (available, not currently used in active tools)
deep-translator ≥1.11.0 Translation fallback (available; Gemini used instead)
pytesseract ≥0.3.10 Tesseract OCR (available; Gemini Vision used instead)
rank_bm25 ≥0.2.2 BM25 search (available for future local search features)

Recommended External Resources for Legal Research

Since this project does not contain a complete database of Indian law, the following authoritative sources should be used alongside it:

Resource URL What It Has
India Code https://www.indiacode.nic.in All Central Acts, official text, amendments
Legislative Department https://legislative.gov.in Bills, acts, statutory rules
Supreme Court of India https://main.sci.gov.in SCI judgments, cause lists, rules
eCourts Services https://services.ecourts.gov.in District/HC cause lists, case status
Indian Kanoon https://indiankanoon.org Free search of SCI/HC/Tribunal judgments
Manupatra / SCC Online (subscription) Comprehensive case law + commentaries
Law Commission of India https://lawcommissionofindia.nic.in Reports, recommendations
MCA21 https://www.mca.gov.in Company law filings, NCLT orders

This README was generated after a full code audit of the project. All claims about what is and is not present in the codebase are based on direct inspection of the source files and data files.

About

Streamlit workbench of ten tools for Indian legal practice: case law research via Indian Kanoon, pleading drafting, judgment summarisation, OCR translation and docket tracking.

litigat-ai.streamlit.app/

Topics

document-draftingecourtsgeminigroqindian-kanoonindian-lawlegal-ailegal-nlplegal-researchlegal-techlitigationllmmultilingualocropenaistreamlit

Resources

Readme
Activity

Stars

1 star

Watchers

0 watching

Forks

0 forks
Report repository

Releases

Packages

Contributors

Languages