Welcome to the Marcura AI CCR
Multi-agent applications
Operators Workbench AI Agents
Description:
When chartering managers hand over a fixture to an operator, they manually fill out a handover template by extracting
information from charter-party (CP) documents and emails. This process is time-consuming and prone to errors.
The goal is to automate the template-filling process, making Operators Workbench a reference point for operators.
This supports the broader product objective of establishing Operators Workbench as the go-to solution for operators managing
their voyages.
Team: PortLog
Repository: https://github.com/Marcura/owb-ai
Test Url: https://handover-demo.livelycoast-d4073e5f.uksouth.azurecontainerapps.io/
Techstack:
- Langchain + LangGraph
- Pydantic for Schema validation
- OpenAI models through Azure AI Foundry
- Serverless Azure Functions for deployment
Tech Components: We have made some components that may be reused in other projects
- Strike-through text detection, which is not handled by standard OCRs. Currently only supports well-formatted text (i.e. non-scanned documents)
- Header and footer detection, which is useful for removing irrelevant information from documents
DA-Desk AI Tariff Screener
Description:
In the maritime and shipping industry, Proforma Disbursement Account (PDA) calculations are crucial for estimating port call
expenses such as port dues, pilotage dues, towage dues, light dues, anchorage dues, etc. These calculations are traditionally
manual, requiring users to extract tariff rules from tariff documents, apply formulas, and factor in vessel parameters. Given the
complexity and variability of tariffs across ports and countries, automating this process with AI presents significant challenges.
Our goal is to build an AI-driven PDA tariff automation system that can:
- Identify relevant documents needed for tariff calculations.
- Extract relevant sections, tables, and formulas to structure them meaningfully.
- Understand the tariff calculation logic and represent it in JSON format.
- Determine parameters required for cost computations.
- Integrate with an external function to execute the calculations.
- Compare computed values against actual DA values for validation.
- Use historical DAs to estimate missing parameters like “days alongside”.
Team: DA-Desk
Repository: https://github.com/Marcura/dadesk-module-ai
Test Url: https://app-backend-6wwxz2q3ho5bo.azurewebsites.net
ShipServ - Invoice Parser
Description:
Put description here
Team: ShipServ
Repository:
Test Url:
Marcura Claims - Clause Search
Description: The goal is to provide Claims Hub internal team the ability to upload CP and recaps and basis a preset selection identifying specific clauses quickly, with a document highlight as a reference. This is a stepping stone into automating CP Hub input
Team: Marcura Claims (Michal Szewczuk / Tiago Bonamigo)
Repositories:
- Experiments: https://github.com/Marcura/datafactory-research-cps
- Production version: https://github.com/Marcura/ch-clause-search-engine
Test Url: https://claims.cuat.marcura.com/calculations
Shipster AI Agents
Description:
Shipster’s vision is to build vertical AI agents for the shipping industry: an agentic tool that is optimized for shipping workflows, and can handle the day-to-day administrative burden of an operator while also pro-actively scanning for and mitigating downside risk or finding favorable opportunities.
As the first step shipster focused on building out a set of tools to accomplish a number of document based tasks that until now are still mostly manual, for example extracting information from a B/L to be inserted into an LOI, comparing MRs to B/Ls, comparing time charter hire statements, etc.
These tools are immediately useful to operators but also form the building blocks for a more autonomous AI agent. The next step is therefore to integrate with email and VMSs, so that we can pro-actively trigger and handle workflows in the background and present ready responses to the agent for confirmation. An AI operations agent will trigger based on e.g. an incoming email from the master, search and find relevant information, use specific tools to accomplish specific tasks (e.g. generate an LOI from a template), and present actionable results to the end user such as a draft email response including attachments.
Currently we are rewriting our old solution to brand new version to be integrated into Marcura (deadline end of April). You can follow the work in our repos.
Things that we may be able to help you out with
- how to use pydantic schemas to extract reliable structured outputs from unstructured documents
- how to use langsmith to gain observability over your llm inputs and outputs
- how to build evals for structured extraction tasks
- how to build an annotation tool using fasthtml for quickly annotating expected outputs for a given input
Tech stack
- frontend: typescript, react, next.js, shadcn
- backend: python, fastapi, pydantic, instructor, openai/gemini, fasthtml
- infra: azure, app containers
Team:
Oege Dijk, Ivana Ivancic, Samantha Swift, Kush Gaikwad, Jaswir Raghoe, Mirta Germovsek
Repositories:
https://github.com/Marcura/shipster-backend
https://github.com/Marcura/shipster-frontend
https://github.com/Marcura/shipster-infra
Test Url: https://staging.app.shipster.club/ Please reach out to the team for internal credentials.