Previously, we developed a paperless life raft inspection sheet system for Swan Net Gundry. This software project was developed in a largely traditional manner, however leveraged agentic coding for most of the heavy lifting of writing code, and also as a technical advisor to the project manager.
SNG Paperless Life raft Inspection: https://github.com/spycebot/sng-paperless-liferaft-inspection Swan Net Gundry: https://sng.ie
National and international law requires that sea-going vessels, such as fishing boats, cargo ships, and survey ships, be equipped with life rafts. In most cases, these rafts are made of natural rubber, and are inflated by means of CO2 cylinders when deployed in an emergency. Life rafts also contain an emergency pack, which in turn contains items such as a first aid kit, emergency food and water, signal flares, heliograph for signalling other ships, ores, a puncture repair kit, and even fishing tackle.
Most life rafts are inspected and certified on a yearly basis. When inspecting life rafts for certification, SNG technicians complete the life raft inspection sheet, which serves as both a checklist of the inspection, and a permanent record of test values, record that each element of the emergency pack is in date as stipulated by law, record of the emergency equipment changed, and similar technical details. SNG keeps these in storage for a minimum of seven (7) years.
Regulation 20 - General requirements for life rafts: https://www.imorules.com/GUID-C2874BB2-AD2F-40D3-B16E-086881092340.html
The goal of the Paperless Life Raft Inspection Sheet was to provide a digital system that would retain the functionality of the paper-based inspection sheet, while bringing additional functionally such as the ability to search past records. Further objectives included the ability to report on specified aspects of the life rafts not captured in the ERP system, and have the data backed-up to secure physical locations.
The project began when I made an oral pitch to top management. After the idea received the green light, I followed up by drafting and submitting a formal written proposal. Once top management approved this written proposal, the proposal was fed to Claude Code as the high level design document.
Working with Claude.ai and Claude Code, the high level design was worked out into a formal technical specification: For the initial (as yet 'proof of concept') stages, we settled on Python Flask for the application logic, SQLite for the database, Flask-SQLAlchemy for the database interface, and PyTest for unit and regression testing. Importantly, Claude was careful to advocate for technical design decisions that would lay the groundwork for future extensibility such as ERP integration.
Python Flask: https://flask.palletsprojects.com/en/stable/
In addition to using the web interface to complete one life raft inspection at a time, the application included functionality to bulk upload data via a CSV / Excel spreadsheet. Likewise, data export in the first instance was by means of CSV (comma separated values). The editing of data already committed to the database is performed via a custom built TUI (terminal user interface), which was built on top of Python's Textual library.
Textual: https://textual.textualize.io/
The project proceeded in Agile-style sprints, each iteration adding a new layer of functionality and complexity. Upon completion of all core functionality as designed, the application was able to record the progress of individual life raft inspections, manage the stored data, and provide reports.
I would like to note that in the process of developing this application, many inconsistencies in the original paper format were uncovered. The way the tables on the paper sheet were laid out assumed a flow of execution that was easy for a human to follow, but difficult for a machine to follow without question.
When human operators work with paper forms, they make a large number of determinations 'on the fly', assumptions about what the form is asking for, and don't spend much time questioning the details of the form itself. However, when working with an AI agent to generate designs or code, each of these determinations should be well thought through: A quick decision now may (and most likely will) lead to design constraints in the future, even unwanted ones.
This having been said, in the next phase of development, I would like to move away from reproducing the paper inspection sheet exactly, and move toward a more well rationalised inspection work flow. As changes to the schema of the underlying database will generate a great deal of project friction, they will necessarily be kept to a minimum. OR, having demonstrated the viability of the project and having obtained the backing of management, perhaps now is the time to migrate to a data schema that will better support ongoing development???
With this in mind, I have worked out the major stages of a life raft inspection to be as follows:
In this initial, 'proof-of-concept' phase of software development, the nice-to-have functionality was of two types: (1) Integration with SNG's ERP and accounts management system, and (2) Bulk scanning and optical character recognition (OCR) of legacy paper inspection sheets.
Integration with SNG's ERP (enterprise resource planning) system was regarded as technically feasible by means of the REST API provided by the ERP package, but not accomplished due to internal administrative hurtles which could not be surmounted within the initial time frame. Additionally, ERP integration of the paperless life raft inspection sheet application took a back seat to day-to-day revenue focused business operations.
Although accomplished on a small scale, OCR of legacy paper inspection sheets could not not be accomplished on a bulk scale in an efficient manner without additional investment in either GPU hardware, or the ringfencing of funds for cloud computing. Utilisation of cloud computing resources was also deemed to impose the need for additional consideration of the geographic location of business data. Using local CPU only, modern OCR models required about 13 minutes per page, and yielded accuracy in the range of 80% for hand-written text.
The next step in the paperless life raft inspection sheet project is the adoption of WiFi enabled android tablets. The Samsung Galaxy Tab A11 tablets were purchased from a national retailer for less than older models could be purchased from refurbished tablet providers, and under a much longer warranty. The 8.7" screen size means the screen will be large enough for easy reading, but small enough to not encumber technical inspection work operations. The paperless life raft inspection sheet application is served over the secured local WiFi network, and the tablets access the application by means of their web browser.
Following a review by top management of the development work done so far, further development of the paperless life raft inspection sheet application is set to continue.
The coming series of development sprints will include: (1) a landing page from which each sub-application is accessed, (2) a life jacket paperless inspection sheet application, (3) automatic printing of the life jacket certificate, (4) a calendar to track technician certification and qualification.
We will develop an application for tracking the details of life jacket inspections that follows along the same lines as the life raft inspection sheet application. Likewise, it will be hosted in-house, and served over the secured local Wifi network.
The format of the user interface for this application will integrate both the paper format developed by SNG internally for its own record keeping, as well as the paper-based inspection sheet format recommended by SNG's life jacket supplier, Mullion.
Mullion Personal Flotation Devices: https://mullion-pfd.com/en/life-jackets
As life jacket certificates are SNG's own proprietary format, they can an will be generated automatically, drawing on inspection details in the database, and printed ready for the technician's signature and the company seal.
The final goal of this phase of development will be a calendar that clearly displays the renewal dates for each inspection technician's own certifications. Each technician must maintain a number of qualification certifications. As well, the number of required certifications is set to grow as SNG's safety business grows.