AI for Maritime Safety Intelligence Platform
Budget: $250 – $750 USD
AI / LLM Engineer Needed – Maritime Safety Intelligence Platform (OceanLens)
Project Overview
I am building , a maritime safety intelligence platform that collects and analyzes marine accident investigation reports from authorities and investigation bodies.
The data collection pipeline is already developed and operational. Reports are downloaded, stored, and indexed in a structured database.
I am now looking for an experienced AI / LLM Engineerto build the intelligence layer of the platform.
Objective
Develop an AI-powered extraction system that converts maritime accident investigation reports (PDFs) into structured, validated data suitable for storage in a relational database.
The system must extract factual information only. It must never invent or hallucinate information.
Scope of Work
1. LLM-Based Information Extraction
The AI should extract information such as:
* Incident title
* Incident type
* Date
* Vessel information
* Location
* Weather conditions
* Equipment involved
* Sequence of events
* Immediate causes
* Root causes
* Contributing factors
* Human factors
* Technical failures
* Environmental factors
* Regulatory issues
* Injuries
* Fatalities
* Pollution
* Property damage
* Lessons learned
* Corrective actions
* Safety recommendations
* Keywords
* Confidence score for each extracted field
Output should be structured JSON matching the existing database schema.
2. Rule Engine
Develop a transparent rule engine that calculates:
* Severity
* Likelihood
* Risk Score
Important:
The LLM should not generate these values.
Instead, the rule engine should calculate them using predefined business rules based on the extracted facts (fatalities, injuries, pollution, vessel damage, operational impact, etc.).
The rules must be configurable and easy to modify.
3. Validation
The system should:
* Handle missing information correctly
* Leave unsupported fields blank
* Never fabricate information
* Provide confidence scores
* Generate logs for extraction errors
4. Integration
Integrate with the existing Python pipeline.
The solution should:
* Process downloaded PDFs
* Produce structured JSON
* Store results in the existing database
* Support batch processing
* Be modular and maintainable
Technical Requirements
Preferred experience with:
* Python
* Large Language Models (OpenAI, Anthropic, Gemini, or open-source models)
* Prompt engineering
* PDF parsing
* OCR (if required)
* JSON schema generation
* FastAPI (preferred)
* PostgreSQL or SQLite
* Git
Deliverables
* Complete AI extraction module
* Rule engine
* API or processing module
* Documentation
* Source code
* Installation instructions
* Sample outputs
* Unit tests
Important
This is not a chatbot project.
This is not a summarization project.
The objective is to build a reliable information extraction system for maritime accident reports with transparent, auditable rule-based risk scoring.
If you have built similar document intelligence, knowledge extraction, or LLM data extraction systems, please include examples in your proposal.
Please also explain:
* Which LLM(s) you recommend and why
* How you would minimize hallucinations
* How you would evaluate extraction accuracy
Project Overview
I am building , a maritime safety intelligence platform that collects and analyzes marine accident investigation reports from authorities and investigation bodies.
The data collection pipeline is already developed and operational. Reports are downloaded, stored, and indexed in a structured database.
I am now looking for an experienced AI / LLM Engineerto build the intelligence layer of the platform.
Objective
Develop an AI-powered extraction system that converts maritime accident investigation reports (PDFs) into structured, validated data suitable for storage in a relational database.
The system must extract factual information only. It must never invent or hallucinate information.
Scope of Work
1. LLM-Based Information Extraction
The AI should extract information such as:
* Incident title
* Incident type
* Date
* Vessel information
* Location
* Weather conditions
* Equipment involved
* Sequence of events
* Immediate causes
* Root causes
* Contributing factors
* Human factors
* Technical failures
* Environmental factors
* Regulatory issues
* Injuries
* Fatalities
* Pollution
* Property damage
* Lessons learned
* Corrective actions
* Safety recommendations
* Keywords
* Confidence score for each extracted field
Output should be structured JSON matching the existing database schema.
2. Rule Engine
Develop a transparent rule engine that calculates:
* Severity
* Likelihood
* Risk Score
Important:
The LLM should not generate these values.
Instead, the rule engine should calculate them using predefined business rules based on the extracted facts (fatalities, injuries, pollution, vessel damage, operational impact, etc.).
The rules must be configurable and easy to modify.
3. Validation
The system should:
* Handle missing information correctly
* Leave unsupported fields blank
* Never fabricate information
* Provide confidence scores
* Generate logs for extraction errors
4. Integration
Integrate with the existing Python pipeline.
The solution should:
* Process downloaded PDFs
* Produce structured JSON
* Store results in the existing database
* Support batch processing
* Be modular and maintainable
Technical Requirements
Preferred experience with:
* Python
* Large Language Models (OpenAI, Anthropic, Gemini, or open-source models)
* Prompt engineering
* PDF parsing
* OCR (if required)
* JSON schema generation
* FastAPI (preferred)
* PostgreSQL or SQLite
* Git
Deliverables
* Complete AI extraction module
* Rule engine
* API or processing module
* Documentation
* Source code
* Installation instructions
* Sample outputs
* Unit tests
Important
This is not a chatbot project.
This is not a summarization project.
The objective is to build a reliable information extraction system for maritime accident reports with transparent, auditable rule-based risk scoring.
If you have built similar document intelligence, knowledge extraction, or LLM data extraction systems, please include examples in your proposal.
Please also explain:
* Which LLM(s) you recommend and why
* How you would minimize hallucinations
* How you would evaluate extraction accuracy