A hobbyist put a circuit-checking AI helper inside a Discord chat

The bench problem

Anyone who has debugged a circuit knows the moment. You have oscilloscope probes in one hand and multimeter leads in the other, and suddenly you need to recompute a voltage divider or pick standard resistor values. Stopping to open simulation software or dig through PDFs breaks your concentration.

As dev.to reported, one developer decided the fix was an assistant living in the chat where their group already trades ideas late at night: Discord. The result is Amply, an AI helper for quick sanity checks on electronics work.

The author names two reasons ordinary AI bots disappoint in a technical channel. First, Discord cannot display LaTeX, a notation used to typeset math, so equations show up as clutter such as raw fraction code. Second, large multimodal models (ones that handle images and text) can be slow, eat through usage limits, and are more than a quick circuit question needs.

How Amply is wired up

The glue is Switch, which the author describes as an open-source tool that connects AI agents to team chat apps including Discord, Slack and Microsoft Teams. Instead of writing a custom Discord bot from scratch, you treat chat channels as shared rooms where an agent can join the conversation like a human teammate. It runs on your own machine and uses Server-Sent Events, a standard way for a program to stream updates, to communicate.

Setup, the author says, took under fifteen minutes using the Switch Console desktop app. They made a local workspace and linked their Discord server through a messaging connector.

For the brains, they chose Google's lightweight gemini-3.5-flash-lite model through an API (a connection that lets one program call another service). The stated reasons were speed, reliability and avoiding rate limits, which are caps on how many requests you can send.

Teaching the bot to write for Discord

The most transferable part is the instruction sheet, known as a system prompt, that tells the model how to behave. Because Discord cannot render LaTeX, the author banned raw delimiters like $$ and \frac. Formulas must use clean Unicode symbols instead, such as ×, ÷, ≈, Ω, µF and π.

Other rules ask for calculations broken into steps with tidy indentation. The bot is also told to recommend commercial component values from the E12 and E24 series, which are standard sets of resistor and capacitor values actually sold in shops, rather than impractical decimals.

None of this is complicated. It is a good example of designing around a platform's limits with plain-language instructions rather than extra software.

What it did in testing

The group tried Amply in their #general channel on realistic tasks. The first was an inverting amplifier, a common op-amp (operational amplifier) circuit that flips the sign of a signal, with a gain of -10. Amply picked 10 kΩ for the input resistor and 100 kΩ for the feedback resistor, then showed that an input of 0.5V gives an output of -5.0 V. It also noted the inverted polarity and advised keeping supply rails above ±7V to avoid clipping, where the output gets flattened at its limits.

The second test used a Sallen-Key low-pass filter, a popular active filter that passes low frequencies and cuts high ones. With R1 and R2 at 10k and C1 and C2 at 10nF, Amply worked out a cutoff of about 1,591.5 Hz, roughly 1.59 kHz.

It also explained why a simulation might show peaking. In equal-component versions of this design, the sharpness of the response, called Q, depends on amplifier gain through Q = 1 / (3 - A). As the gain nears 3, Q spikes, which can cause sharp peaks or oscillation near the cutoff.

Lessons and limits

The author's takeaways are simple. Meeting a team in the app it already uses keeps work collaborative and cuts context switching. Platform limits can be handled in the prompt. And a targeted prompt with a fast, small model can serve a specific domain better than a huge, slow one.

Keep the evidence in proportion. This is a personal build write-up with a handful of example questions, not a benchmark. The two examples shown are textbook-style problems, and the post does not report how Amply handles messy, ambiguous real circuits or whether it ever gets things wrong.

That is why the sensible habit applies: treat the bot as a fast second pair of eyes, and check anything that could damage parts or a grade against a textbook or a simulator.