From ../kicad-walkthrough.md,
../schematic-notes.md and
../documentation.md §6–10.
Q1. [Intermediate] 🔥 Walk me through how you actually built this schematic.
Strong answer covers the phase order and why it’s that order:
- Project setup — sheets, title block, libraries.
- Place the ESP32-C6 first — it’s the hub, and every other block’s pin assignment is decided by
what the MCU has available. - Power section next — because every other block needs a rail to connect to, and the rails
determine which parts are even viable. - Peripherals — LoRa, Ethernet, cellular, then the USB/SD/button/LED/SWD section.
- Final ERC and cleanup.
The specific tactic worth naming: the PoE controller is a 47-pin part, so it was placed early
and wired last — get the footprint and the sheet layout settled while the surrounding context is
simple, then do the tedious wiring when everything it connects to already exists.
The mental model to state: the schematic isn’t drawn left-to-right; it’s built hub outward,
power first.
Q2. [Intermediate] 🔥 What is ERC and what does it actually catch?
Strong answer covers: Electrical Rules Check validates the schematic’s connectivity — unconnected
pins, outputs driven against each other, inputs with no driver, and power nets with no source. It is
not a design review: it can’t tell you a resistor is the wrong value, a supply is out of range,
or that you’ve connected TX to TX. It catches the mechanical mistakes so your attention is free for
the ones that need judgement.
The habit: run it continuously, not once at the end. A clean ERC after every section means each
new error belongs to the thing you just drew.
Q3. [Advanced] 🔥 Explain PWR_FLAG. When does a net need one and when doesn’t it?
Strong answer covers: ERC insists every power net be driven by something. Regulator outputs are
declared as power outputs, so a rail fed by an LDO’s VOUT already has a driver. But a net that enters
the board from outside — from a USB cable, or from a source you haven’t wired yet — has no pin in the
schematic claiming to drive it, so ERC reports “power input not driven.” PWR_FLAG is the annotation
that says “this is externally sourced, trust me.”
The project’s rule, stated precisely: only on externally-sourced nets — USB_VBUS (comes from
the cable) and POE_5V (because the PPS23730 isn’t wired yet). Nets fed by an LDO’s VOUT pin do
not need one.
Why this is a good answer: the wrong instinct is to spray PWR_FLAG on every rail until ERC goes
quiet, which suppresses exactly the errors that would have told you a rail has no source. Knowing
when not to use it is the point.
Q4. [Intermediate] How do you know your pin assignments are actually legal?
Strong answer covers: GPIO on an ESP32 is not uniform — some pins are strapping pins that
determine boot mode and must be at a known level at reset; some are input-only; some are consumed by
the module’s own flash and must never be used; and USB and JTAG functions occupy specific pins. So
assignment is a constraint-satisfaction exercise against the datasheet’s pin-function table, not a
“pick a free pin” exercise. The failure mode is a board that behaves perfectly until a peripheral
holds a strapping pin low at power-on and it silently boots into download mode.
Q5. [Intermediate] 🔥 What’s the difference between the schematic and the layout, and what’s still ahead of you?
Strong answer covers: the schematic captures connectivity — what connects to what. The layout
is physics — where components sit, how wide and how long the traces are, stack-up, impedance
control, thermal relief, and every EMC consequence. A netlist that’s perfect in schematic can produce
a board that doesn’t work, and almost every hard problem on a multi-radio board (Q9 in
02-power-and-rf.md) is a layout problem.
Be honest about status: the schematic work — power, MCU, LoRa, Ethernet, cellular, and the
USB/SD/button/LED/SWD section — is what’s done, along with component selection and the design-decision
record. Layout, DRC, gerber generation, and a fabricated board are outstanding. Saying that plainly is
much stronger than implying a finished product; interviewers respect a precise status.
Q6. [Advanced] 🔥 Which decision in this project are you least sure about?
Strong answer covers — pick a real one and reason both ways:
- The LDO for 3.3 V. Simple and quiet, but it burns ~0.85 W at 500 mA. If the real load turns out
higher, that’s a thermal problem on a small board and a buck converter becomes necessary — with
switching noise arriving next to three radios. - A shared 16 MHz SPI bus across radio, Ethernet and SD. It’s clean and pin-efficient, but the SD
card can hold the bus during internal operations, and that latency lands on a radio that has timing
expectations. A second SPI peripheral for the SD card is the fix if it bites. - Antenna placement and isolation, which is genuinely unresolved until there’s a fabricated board
to measure. That’s the honest answer to “what worries you”, and it’s not something a schematic can
settle.
Q7. [Intermediate] 🔥 Why does a hardware project need this much documentation?
Strong answer covers: because in hardware, the cost of being wrong is a respin — weeks and
real money, not a redeploy. A written decision record (why 868 MHz, why a module, why the cellular
rail is separate, why no external flash) is what stops the same question being re-litigated from
memory three months later, and it’s what a reviewer or a manufacturer needs in order to check your
reasoning rather than just your netlist.
The second reason: a schematic records what you did and never why. Six months on, “why is
there a 49.9 Ω resistor on the centre tap?” is answerable from a datasheet reference in your notes and
essentially unanswerable from the schematic alone.
Q8. [Beginner] You’d never used KiCad before. How did you learn it?
Strong answer covers: by building the actual board rather than by working through tutorials —
with a written walkthrough capturing the phases, plus a keyboard-shortcut sheet, so the tool stopped
being the bottleneck. The transferable point: for a tool you’ll use for one project, learn it against
the real artefact and write down what you had to look up; for a tool you’ll use for years, invest in
fluency separately. Naming which situation you were in is the answer.
Q9. [Advanced] What’s the riskiest part of this design, and how would you de-risk it before committing to fabrication?
Strong answer covers, in order:
- RF performance — LoRa range and multi-radio coexistence can’t be verified from a schematic.
De-risk with dev-kit prototyping: an ESP32-C6 board plus an SX1262 module plus a SIM7080G breakout
on a bench, proving the firmware and the coexistence story before laying out a single trace. - The PoE front end — the most complex part, currently unwired, and the section with safety and
isolation implications. De-risk by prototyping against a reference design and a real 802.3af
switch. - Power budget under simultaneous worst case — every radio active at once. That’s an arithmetic
exercise against datasheet peak currents, and it’s cheap to do before it’s expensive to discover.
The framing: buy a dev kit for every risky subsystem and prove the software before you spin a
board. The board is the last step, not the first.
Q10. [Intermediate] 🔥 What did designing hardware teach you that software hadn’t?
Strong answer covers — pick one and go deep:
- Irreversibility changes the process. You can’t hotfix a PCB. That’s what makes the documented
decision record, the DNP eSIM footprint, and the “place it now, wire it last” discipline rational
rather than fussy — the same reasoning as designing around irreversible actions in software, but
enforced by physics instead of by judgement. - The datasheet is the API, and it’s non-negotiable. A 49.9 Ω centre-tap resistor isn’t a
preference; it’s a stated requirement, and there’s no runtime that will forgive getting it wrong. - Constraints propagate physically. Choosing a cellular modem forced a second regulator, which
forced a thermal question, which constrained the layout. Software dependencies rarely couple that
tightly.
Q11. [Beginner] Explain this board to a non-technical interviewer.
Strong answer covers: it’s a translator box for connected devices. Sensors in a building or a
field speak short-range low-power radio protocols and can’t reach the internet on their own; this
board listens to all of the common ones, then forwards the data over whichever internet connection
is available — a network cable, Wi-Fi, or a mobile network if there’s nothing else. The engineering
challenge is fitting four radios and three power sources onto one board without them interfering with
each other.
Software engineering
Infrastructure & security
Computer science
Hardware & engineering