GPT-6 Astra in Action: How Someone Without an Engineering Background Used AI to Build Hardware

How Someone Without an Engineering Background Can Launch a Hardware Project with AI

Now that AI-assisted programming has become the norm, a more interesting question has emerged: Can AI help people without professional backgrounds tackle real physical hardware engineering?

A Reddit user shared his complete experience using GPT-6 Astra + Codex to develop a hardware project called the "Jornada P6." The author is upfront that he is not an electrical engineer and has little background in electronics or mechanical engineering — and that's precisely the most important premise of the whole story.

His goal was to build a modern version of the late-90s HP Jornada handheld PC: a "small, keyboard-first, distraction-free writing computer." The idea itself isn't new, but what moved it from concept render to physical reality was the role AI played across the entire engineering reasoning chain.

The Jornada P6 Hardware Spec: Vintage Keyboard + E-Ink + Custom Controller

The Jornada P6 hardware design blends retro and modern:

• Original HP Jornada 688 keyboard — 25-year-old hardware requiring reverse engineering
• Detachable BOOX P6 e-ink device — serving as the display
• Custom nRF52840-based controller — handling keyboard signals
• BLE HID for everyday wireless use, USB for charging, firmware, debugging, and fallback HID
• A fully new enclosure, hinge, and internal mechanical structure

What the author could handle was clear: defining the product form, disassembling and measuring, soldering components, printing prototypes, and judging whether a part would fit. But the real engineering challenges — how to reverse-engineer the keyboard matrix, how to handle FFC connections, how HID works, MCU architecture selection, PCB planning, battery and charging architecture, hinge load paths, magnetic versus shear retention, internal space layout, serviceability design — he had zero hands-on experience with any of it.

The AI-Driven Engineering Loop: Define, Experiment, Feedback, Iterate

What makes this project most noteworthy is the stable working loop the author developed with GPT-6 Astra:

I describe a goal → Astra figures out what we need to know → it designs an experiment → I do the physical measurement/test → I feed back results → Astra interprets the results → Codex updates the project/design → we identify the next unknown → repeat.

Take disassembling the original Jornada 688 keyboard as an example. The author had no idea what the keyboard matrix structure looked like. Astra helped him reason through the FFC ribbon cable, establish a testing methodology, safely probe the vintage keyboard, and interpret the electrical signal results — gradually transforming a mystery piece of old hardware into a component that could be driven by a modern wireless controller.

Mechanical design followed the same pattern. The project evolved from a "cool concept render" into a real conversation about actual dimensions, hinge geometry, load paths, detachable display retention, PCB/battery volume, USB-C port placement, fasteners, service covers, and more. The author says this transition — from "cool picture to the next physical experiment" — is what impressed him most about Astra.

The Power of Open Source: Standing on the Shoulders of Giants

To the author's credit, he repeatedly emphasizes that this project didn't emerge from a vacuum — it draws heavily on prior work from the open-source community. He specifically credits several reference projects:

• Un Kyu Lee's Micro Journal project (especially PDA Terminal), which provided the primary inspiration for integrating a mobile device with a keyboard in a compact clamshell form factor
• Wujek_k's Jornada nostalgia handheld (Thingiverse 5590156), demonstrating how to reshape the old Jornada form around modern hardware
• Jornada Nostalgia Handheld v.2 (Thingiverse 6938611), which pushed further on mechanical architecture and internal space layout
• Jornada keyboard replacement (Thingiverse 6966529)
• 38911-BASIC-BYTES-FREE's USB-Jornada keyboard board and RasmusB's USB keyboard adapter, key references for converting the old keyboard to USB HID
• PH Design's PH60, whose open mechanical and wireless keyboard design provided a reference for the nRF52840 architecture

The author acknowledges: "Being able to examine how other people solved hinges, keyboard interfaces, battery packaging, printable enclosures, and control electronics dramatically reduced the amount of blind guessing." This reveals an important truth — AI doesn't generate solutions in isolation; its value lies in helping users efficiently understand, integrate, and adapt the accumulated engineering wisdom of the human community.

The Risk of AI Hallucinations in Hardware: More Dangerous Than Software Bugs

The most valuable insight in this writeup is the author's clear-eyed awareness of the "hallucination" risk AI poses in hardware engineering:

If a model fabricates a CSS value, maybe the page looks slightly wrong. If it fabricates a mechanical dimension, your parts might not fit at all.

AI errors in software can usually be corrected at low cost, but errors in hardware mean wasted money on materials and parts that don't work. To address this, the author established an evidence hierarchy for the entire project:

1. Measured facts
2. Verified datasheets
3. Reference evidence
4. Engineering assumptions
5. TBD

The core rule: if a piece of data has not been physically measured or lacks reliable documentation, Astra/Codex is not permitted to silently elevate it to fact. This rule became critically important throughout the project — it's essentially a "hallucination guardrail" for AI-assisted hardware development, and it's well worth borrowing for anyone attempting similar work.

Current Project Status and Potential Failure Points

The author maintains a refreshingly honest view of where things stand. Work completed so far:

• Real Jornada 688 keyboard disassembled and measured
• FFC ribbon cable characterized
• Keyboard matrix reverse engineering in progress
• BOOX P6 HID interaction tested
• BLE/USB system architecture mostly defined
• Multiple mechanical design candidates produced and reviewed

Upcoming milestones include: completing the full keyboard matrix map, freezing the keyboard interface, finalizing the nRF52840 electronics design, finalizing battery/charging/USB-C layout, printing the first real chassis, physical hinge testing, and building the first integrated EVT prototype.

The author also lists potential failure points with refreshing candor: the hinge might not work, the internal space might be too tight, the keyboard electronics might reveal surprises, and the first print will almost certainly expose mistakes. He explicitly states he is "absolutely not publishing this as a success story."

AI Is Changing Who Gets to Make Things

The author articulates what makes this experiment most fascinating: what's truly compelling isn't that AI can answer electronics questions or generate CAD code — it's that a person with no rigorous engineering background can now sustain a continuous, coherent conversation across the entire chain of product design → research → reverse engineering → electronics → mechanical design → firmware → testing → documentation, and push the same physical project forward.

He recalls that a few years ago, the best he could have done was post a nice concept render online and stop there. Now, on his desk sits a disassembled 25-year-old keyboard with wires running out of it, a matrix being mapped one connection at a time, mechanical designs under review, and a first printable prototype gradually taking shape.

This may be one of the most profound impacts of the new generation of AI models — it doesn't just make professionals more efficient; it lowers the barrier to cross-domain fabrication, fundamentally changing who is "qualified" to build physical things. When AI can serve as a patient engineering tutor, experiment designer, and knowledge translator, the once-formidable professional wall separating makers from hardware engineering is quietly being reshaped.