Wrapping Up the Pilot Experience

Previous Posts

• Week 6: One Task at a Time

• Week 5: Building on a Strong Foundation

• Week 4: Choosing a Clear Direction

• Week 3: Designing for Different Users

• Week 2: Organization Improves AI Results

• Week 1: The Importance of Staying Organized

Goal

• Finalize pilot version of RuffRide
• Begin working on companion dashboard

Workflow

This week was another successful week for me. Anthropic keeps extending the date by which Fable, their most powerful model, is removed from subscription plans, allowing me to continue benefiting from its power.

Here’s a breakdown of what I accomplished this week as I finish up RuffRide. 

• More Terrier Layers
  • Pressure
  • Visibility
  • Precip
  • Cloud Cover
• Plotted METAR plots as wind barbs
• PIREPS (Pilot Reports)
• Lightning (from GOES GLM)
• Flight Plan Tab

The flight plan tab underwent the most changes. There are now 4 sub-tabs that open up under Flight Plan. These are:

• Plan
  • Where the user uploads or manually inputs their flight plan. This is highly standardized and includes many information boxes, although the route itself is enough to generate the FP.
• Departure / Arrival
  • The departure and arrival tabs show detailed information about the departure and arrival airports, depending on the FP entered. On the left, the user can switch between a zoomed-out map view showing the procedures drawn out, and the radar and radar forecast options. The taxi view shows the airport and all live aircraft taxiing on it. The chart view contains all the charts for that specific airport, including the taxi, SIDs, STARs, and various transitions. On the right-hand side, the page is filled with live weather information for the airport.
• In-Flight
  • This section shows the FP drawn out on a map, and it’s also the spot where users will find the vertical 2D turbulence profile. The data is now higher-resolution, and altitude is taken into account. 

Those are the main features of RuffRide, the pilot’s version. There are still small bugs and QOL improvements to make, but my plan is to shift focus and start the physical work for the passenger version this week. I’ll spend the last week or two of this internship refining both final products.

I have a clearer picture of what the companion app will look like. It’s going to be simplistic in data, but heavy on visualizations. Throughout the summer, I’ve found that AI does a good job of working with numbers and data and displaying them. However, my next question is: How good is AI at creative visualization and generation? Turbulence data and WDW products will still factor into calculating flying comfort scores, but this will all be behind the scenes. 

I have a pretty big idea/plan for the companion app that I’m going to tackle. In essence, I’m going to implement a chatbot that answers general questions about aviation weather and turbulence, along with specific details about an upcoming or active flight. The core idea is that they’re the professional who ingests the data from the pilot’s app and conveys it to your everyday passenger. In addition, I want the bot to find you a flight based on your turbulence preferences. These are all just initial ideas, but I feel that if I make this companion app not only AI-built but AI-centered, it could be a massive success. 

Prompts Used

A lot of the prompts I used this week instructed the AI to do online research beforehand to find external data sources. This is how I found the turbulence and lightning data, and then I got all the aviation charts and waypoints/fixes from the FAA. The prompts used to implement have not been much different from previous weeks. 

Lessons Learned

In addition to building and coding, Claude can browse the internet and other relevant sources to gather information and data.