A national aviation regulator's portal (independent concept)

The problem

A national civil aviation regulator publishes the rules, forms and guidance that pilots, engineers and operators depend on. On its current site, that material is hard to use. I audited it and found licensing, the most common task, three clicks deep in a menu of 30+ items arranged by internal department. Documents are served under hashed file names, so neither people nor search engines can tell them apart. The Forms page links each of its 81 files four times. A student pilot or an operator applying for a certificate has to dig to find the right document.

Constraints

• This is an unsolicited, independent concept. It is not the official website and has not been adopted. The regulator makes the final decision, and nothing goes live without its approval.
• The plan calls for hosting inside the country, while the AI, map tiles and flight feeds are outside services. That needs a policy ruling and a personal-data filter.
• Slow mobile networks. The audit set a budget of under 200 KB of JavaScript on content pages.
• The regulator's own IT team would take over six months after launch, so the stack has to stay mainstream and well documented.

What I did

Strategy and brand

I wrote a site audit, a seven-phase roadmap with a soft launch after phase three, and a decision log with 36 recorded decisions. Each one notes what it changes and what risk it creates. The log keeps the existing licensing system as the official system of record and drops online applications from scope, which lowers risk. I kept the existing identity and proposed no new logo, because the official mark carries national symbols.

Design

I reorganized navigation around visitor tasks and moved reporting channels and the exam calendar to where people look for them. The light logo blue fails contrast on white (1.65:1), so I built a color system that keeps it as the signature on navy and uses darker shades (5.64:1 and up) for buttons and links. Licensing journeys became visual pathways, starting with pilot license, aircraft maintenance engineer, foreign license validation and air operator certificate.

Engineering

The current build is a React, TypeScript and Tailwind app with a document library and an in-site reader. A Python script (PyMuPDF, python-docx, Tesseract) extracted text and a table of contents from 424 documents, with OCR for 27 scanned files. The reader shows readable names, an outline and a note that the PDF stays the official version. A live 3D aviation viewer runs on CesiumJS in its own frame. It is built on an MIT-licensed open-source viewer, which I cut down to the civil-aviation parts. It shows live aircraft and aerodrome weather from free public feeds and carries a "Not for operational or navigational use" notice. A small PowerShell server lets anyone launch the prototype on Windows with a double-click. It forwards only to an allowlist of seven data hosts and caches their answers briefly.

AI

I specified an AI assistant that answers from three layers: the documents, a process graph of each licensing workflow, and live context such as exam dates. Its rules are to cite the Part, section or form behind every answer, never decide whether someone qualifies, and hand off to staff. Personal details are removed before anything leaves the server. The plan includes a test set of 200+ real front-desk questions. In the prototype, the assistant runs on the device with voice input and read-aloud, and nothing is sent anywhere.

Outcome

A working prototype and a full set of planning documents. It is an independent concept, not the official website. Nothing is adopted or live, and the regulator has not endorsed it. Usage and impact have not been measured.

What I'd change next time

I'd confirm the hosting provider and the data-licensing budget before polishing visuals. Those two answers decide whether the live viewer can ship at all. I'd also get a named subject-matter reviewer for each pathway early, because the content has to be checked before it can be trusted.