I built a navigation tool that lets you trade time for increased privacy on the road.

01

The camera dots were already on a map. I wanted the route to care about them.

The idea started about a year before the Alpha existed. While working for a cybersecurity client, I started following the criticism around Flock cameras and other automated license plate reader networks. Then I found DeFlock, a crowdsourced map of known camera locations, and started adding two and two.

At the time, DeFlock did not offer navigation. Google Maps could get me from Orlando to Miami, but it could not treat a layer of known camera locations as something I might want to avoid while calculating the route.

Showing camera dots was not the part I was interested in. I wanted the routing system to treat privacy itself as part of the routing decision, starting with known camera exposure instead of assuming the fastest route was automatically the best route.

02

A fast route and a private route are not always the same thing.

Most navigation products optimize for getting there efficiently. Kite adds another question: how much privacy is the driver willing to trade for time?

The first version of that idea focused almost entirely on known automated license plate reader locations. Late in Alpha development, I realized that was too narrow. A route could avoid mapped cameras and still lean heavily on toll roads, creating another privacy tradeoff. Kite now minimizes toll-road use as part of the privacy-oriented routing logic without treating tolls as an absolute ban.

Sometimes the most private route is absurd. Kite still shows it.

On one Miami to Orlando test, the normal trip was roughly three to four hours. An extreme lower-camera option approached nine. I did not want the product quietly deciding that was too inconvenient to show. Kite can make the cost unmistakable. The choice still belongs to the driver.

03

This was for drivers who care about privacy enough to make their own tradeoff.

I designed Kite for privacy-conscious drivers, people familiar with projects like DeFlock, and people who simply do not like the growth of automated license plate reader networks.

That led to two product principles that stayed intact from the beginning.

04

The product depended on data I could verify, but not control.

05

The first version was basically a little text box sitting on top of a map.

The visible interface was never the hard part. Most changes were behavioral. Logic changed less often, but when it changed, it usually reshaped the product.

There were dozens of numbered builds, but many were a one-line fix or a few changed characters. For the design story, I think about the work as seven meaningful product shifts rather than 80-plus "iterations."

06

An international trip is not one trip just because the user typed it that way.

Cross-border behavior consumed a disproportionate share of the build. The early mental model was simple: identify the country, load the right camera data and calculate the route.

That breaks almost immediately.

For Alpha, I bounded continuous road routing to the connected network from Alaska and Canada through Panama, with the Darién Gap as the practical southern end. Remote U.S. road networks such as Puerto Rico and Hawaii are treated as local networks instead of letting the system invent a road across an ocean.

07

Key product decisions

08

Adding Start drive changed what users thought Kite was.

Drive mode was not essential to the first concept. Leaving it out started to feel dishonest about how people actually use map products.

Before pressing Start drive, people judged Kite as a route-comparison tool. The moment the button appeared, the expectation changed. The most common question became: What happens if I miss the turn?

Drive mode eventually gained live location tracking, maneuver progress, ETA and remaining distance, weak-GPS handling, sustained off-route detection, rerouting from the current location, arrival states and best-effort wake lock.

More importantly, a reroute has to preserve the user's Fastest, Balanced or Fewest Known Cameras preference where possible, and it has to re-run the camera verification before making another lower-exposure claim.

09

By the end, the testing question was not "do you get it?" It was "can I make it lie?"

I used a 15-person remote focus group to pressure-test comprehension and expected behavior, then paired that with QC, code review and manual mobile testing. The focus group was useful for gathering different perspectives, but it was not a substitute for real road use.

Once comprehension stopped being the main problem, the test plan got uglier on purpose.

I built a separate qc.html regression harness that hits live services instead of mocked routing. It mixes permanent known-route cases with randomized stress tests for long routes, borders, stale state, camera provenance, duplicate merging, mobile layouts, active Drive mode and failed-reroute preservation.

10

The Alpha exists. The original market gap does not.

Kite reached a working Alpha after roughly a year of part-time work and several concentrated sprints over the final two months.

I then used it to guide a morning and afternoon of errands, about 50 miles of driving in total. Some of the privacy-oriented routes were scenic and indirect, but Kite got me to every destination I asked it to reach.

During that time, DeFlock added native navigation of its own. That means Kite no longer needs to exist because "nobody else does this." Timing solved that part of the problem for me.

I decided to finish anyway. I did not want to leave a year of work half-done, and by then the value of the project had shifted from launching a unique product to learning how to design one.