Skip to content
Enrique Bay ESEN
Projects · Study and 3D model · 2026

The Alfonso Reyes road study

A neighborhood’s exits onto a fast avenue, measured with a drone, simulated and modeled in 3D so the discussion can run on data.

3D render of a four-lane avenue with trees and buildings on both sides
The 3D model of the stretch
Time
Since October 29, 2026. Still under review
Models
Claude
Status
Preliminary

What it is

My neighbors and I had spent years arguing about how to get out of the neighborhood onto Alfonso Reyes avenue. Everyone had an opinion. Nobody had data. So I started measuring.

The result is a preliminary study, with a diagnosis and options in stages. It does not replace a licensed expert’s report and it is still under review, so here I only tell the process.

Blurred aerial view of the avenue, with a box and a speed over each vehicle
The drone footage, with every vehicle detected and blurred

How it was made

  1. The question before the data

    I wrote down the question and five hypotheses before measuring anything. One was uncomfortable: that the most requested fix might not meet the technical criteria.

    Claude · Obsidian

  2. Measure with a drone

    A 135-gram drone, hovering still at 70 meters with the camera pointing straight down, recorded rush hour. A detector trained on aerial images tracked every car to count traffic and work out speeds.

    DJI Neo · Python · OpenCV · YOLO

  3. Standards and evidence

    Four agents researched in parallel: Mexican standards, international evidence and the local context. That left about 60 linked sources.

    Claude

  4. Simulate

    A traffic model calibrated with the measurements. Out of 36 options, 6 made the cut, and each one ran many times: more than 1,300 runs in the latest version.

    SUMO · Python

  5. See it in 3D

    The stretch was modeled on top of the drone’s orthophoto, at 5 centimeters per pixel. The cars in the 3D model are the ones from the simulation, one by one.

    Blender · three.js

  6. A reviewer who did not write it

    A fresh AI session, with no context, reviewed it looking for mistakes. The first pass found 1 serious issue, 14 medium and 17 minor. The latest pass found no serious ones.

    Claude

Models and tools

Claude
Orchestration, analysis, writing and the independent review
DJI Neo
The drone
Python, OpenCV and YOLO
Vehicle detection and tracking in the footage
SUMO
Open-source traffic simulator
Blender and three.js
The 3D model and its browser version

The creative part

Every number carries a label: measured, estimated, simulated, reference or proposed. None was made up to fill a table.

I asked the AI for what I would ask of any serious consultant: tell me where each number comes from. The most serious mistake was caught by a reviewer that had not seen the work.

Time and cost

Time
Since October 29, 2026. Still under review
SUMO, Blender and Python
Open source, no cost
Subscriptions
Claude, which I already pay for

More projects

See all