Charter Party Clause Parser

A layout-aware Python parser for extracting legal clauses from a maritime voyage charter party agreement PDF into structured JSON

This project was developed for the Marcura AI Engineer Challenge

The parser processes Part II of the document (pages 6–39) and extracts clauses in order while preserving:

• clause id
• clause title
• clause text

The implementation uses a deterministic-first parsing pipeline with optional LLM fallback and deterministic post-processing to handle layout inconsistencies in the original document

---

Task Scope

The source document is a voyage charter party agreement containing two main sections:

• Part I — voyage particulars and operational details
• Part II — numbered legal clauses

This solution parses Part II only, as required by the challenge.

For each clause the parser extracts:

• id
• title
• text

Additional constraints from the task:

• clauses must be returned in the same order as in the document
• strike-through text must not be included
• the code must run locally
• API keys must not be committed

---

Output

The parser generates two output files in:

data/output/clauses_flat.json

Clauses grouped by document section clauses_grouped.json

A flat ordered list of clauses

Example:

{
  "id": "5",
  "title": "FREIGHT REMITTANCE",
  "text": "Freight, Dead-freight and Demurrage payable in U.S.D. via T.T."
}

Both outputs preserve the original ordering of clauses.

---

Parsing Approach

The parser uses a deterministic-first extraction strategy.

Pipeline stages include:

• PDF text extraction
• text normalization
• clause identifier detection
• clause grouping
• title extraction
• clause body assembly
• deterministic cleanup
• structured JSON generation

LLM assistance is optional and used only as fallback

For the test of LLM assistance - a key for Antropic API is needed Be sure to take it to the .env.example file before running the pipeline

---

Clause Groups

group_order	section	description
1	Main Charter Party Clauses	core legal clauses
2	Additional Clauses	supplemental clauses
3	Essar Rider Clauses	rider clauses with different formatting

---

Environment Setup

Create environment:

bash scripts/setup_conda_env.sh

Activate:

conda activate charter-parser-2

Install dependencies:

pip install -r requirements-dev.txt

---

Run Parser

PYTHONPATH=src python -m charter_parser.main

or

bash scripts/run_parser.sh

---

Development Tools

pytest  
ruff check .  
mypy src

---

Final Note

The goal of this project is deterministic and reproducible clause extraction despite layout inconsistencies in the original PDF