DIY Mechanic Uses Claude AI and $15 in Parts to Resurrect Failing Jeep's Emissions System
A hobbyist leveraged Anthropic's Claude and OpenAI's Codex to diagnose and repair a 1994 Jeep Cherokee's engine emissions problems, bypassing expensive smog test failures with custom firmware injected through an Arduino microcontroller.
Muhammad Zulali successfully rehabilitated his aging 1994 Jeep Cherokee to meet California's stringent smog test requirements by combining artificial intelligence guidance with modest hardware modifications. The 32-year-old vehicle's primitive engine computer became accessible through a laptop connection via an Arduino Nano board, enabling both diagnostic readings and adjustments to engine parameters.
Zulali's documented effort shows how the Jeep repeatedly failed California's emissions standards by producing excessive nitrogen oxides at specific test speeds. The vehicle initially flunked at 25 miles per hour. Following his attempts to rebalance the oxygen mixture, performance deteriorated further, with the car subsequently failing at both 25 and 50 miles per hour. Each regulatory test cost nearly $100, creating substantial financial pressure to find a solution.
The project's technical foundation relied on Claude identifying that the Jeep's engine operated on a Chrysler SCI bus running at approximately 7,800 baud with inverted electrical polarity compared to standard serial connections. An Arduino Nano handled the necessary signal inversion and bridged the vintage vehicle to contemporary computing equipment. The diagnostic connector on the Cherokee required only three wires from the breadboarded microcontroller to establish communication.
Initial connection succeeded in retrieving basic telemetry including coolant temperature and battery voltage. Zulali expanded the system's capabilities by directing Claude to gather comprehensive engine data during test drives at both 25 and 50 miles per hour, capturing fault codes and monitoring fuel pressure dynamics. The AI processed this information to develop a diagnosis and remediation strategy.
A critical limitation emerged when Zulali discovered the Jeep's main engine control unit lacked reflashable firmware, being just slightly too old to support that capability. Rather than abandoning the approach, he enlisted Codex to write code that could be injected directly into a nearby EEPROM chip, circumventing the firmware restriction.
Claude did most of the diagnostic work and Codex did pretty much all of the memory work. I just did the wiring and the driving.
Muhammad Zulali
Zulali emphasized that accomplishing this repair required neither a professional mechanic nor proprietary Chrysler dealer diagnostic equipment. The total bill of materials came to approximately $15, with the $3 Arduino Nano serving as the project's central component. He has published the work as open-source software under the GPLv3 license in a GitHub repository named xj-scanner, making the approach available to other vehicle owners facing similar challenges.