Every engineering student or hardware tinkerer knows the bench struggle.
You are sitting at an electronics workstation, oscilloscope probes in one hand, multimeter test leads in the other, trying to debug a stubborn analog circuit. Suddenly, you realize you need to recalculate a voltage divider ratio, pick standard E12 resistor values, or check why an active filter is peaking in simulation.
Stopping what you are doing to alt-tab into desktop SPICE software, search through messy PDFs, or manually recompute formulas completely breaks your flow.
Most of our group discussions and late-night project brainstorming happen on Discord. Naturally, having an AI assistant sitting in our team chat to do these quick bench sanity checks sounded ideal. But bringing standard LLM bots into a technical chat channel usually comes with two frustrating roadblocks:
Broken Math Formatting: Discord does not support LaTeX. Ask an ordinary AI for an equation, and you end up with unreadable clutter like \frac{1}{2\pi RC}.
Heavy Latency & High Overhead: Large multimodal models take too long to respond, eat up rate limits, and are complete overkill when you just need quick circuit feedback.
To solve this, I built Amply using Switch, bridging a local, lightweight AI agent directly into our Discord server.
What is Switch?
Switch is an open-source tool designed to connect AI agents to team chat apps like Discord, Slack, and Microsoft Teams. Instead of building boilerplate Discord bots, registering webhook endpoints, or managing websockets, Switch turns your chat channels into shared "rooms."
It runs locally on your machine, handles the agent lifecycle and communication via Server-Sent Events (SSE), and lets your agents participate in discussions just like any human teammate.
How I Set Up Amply
Setting up Amply took under fifteen minutes using the Switch Console desktop app.
I created a dedicated local workspace and connected our Discord server through Switch's messaging connector. For the engine, I paired Switch with Google’s lightweight gemini-3.5-flash-lite model via API. This kept responses fast, reliable, and completely free of rate-limit bottlenecks.
To address the Discord formatting issue, I added strict plain-text and Unicode formatting rules to Amply's system instructions:
Never output raw LaTeX delimiters like $$ or \frac.
Format formulas with clean Unicode symbols (×, ÷, ≈, Ω, µF, π).
Break calculations down step-by-step with clean text indentation.
Always recommend standard commercial component values (E12/E24 series) rather than impractical decimal numbers.
Putting Amply to the Test
Once the agent was active in our #general Discord channel, we ran it through real lab scenarios.
- Inverting Op-Amp Design with Real-World Components Instead of calculating pure theoretical values, we asked Amply to design an inverting amplifier with a voltage gain of -10:
Query: @amply Design an inverting op-amp amplifier with a voltage gain of -10 using standard resistor values. If input voltage Vin is 0.5V, what is Vout?
Amply chose standard, readily available bench resistors:
R_in = 10 kΩ
Rf = 100 kΩ
Gain: Av = -(Rf / R_in) = -(100 kΩ / 10 kΩ) = -10
Output: Vout = Av × Vin = -10 × 0.5 V = -5.0 V
It cleanly laid out the inverted polarity and reminded us to keep supply rails above ±7V to avoid clipping.
- Sallen-Key Active Filter Debugging Next, we asked it to compute the cutoff frequency of an active low-pass filter and troubleshoot AC sweep peaking:
Query: @amply How do I calculate the -3dB cutoff frequency of a Sallen-Key low-pass filter with R1=R2=10k and C1=C2=10nF, and why might my simulation AC sweep show unexpected peaking?
Amply calculated the cutoff frequency cleanly without raw code blocks:
fc = 1 / (2 × π × R × C)
fc = 1 / (2 × 3.1416 × 10,000 × 10 × 10^(-9))
fc ≈ 1,591.5 Hz (~1.59 kHz)
More importantly, it explained why peaking occurs: in equal-component Sallen-Key designs, damping depends directly on amplifier gain via Q = 1 / (3 - A). If the passband gain gets close to 3, Q spikes, leading to sharp gain peaking or oscillations near the cutoff frequency.
What I Learned
Building Amply highlighted a few key lessons about developer agents:
Meeting teams where they already talk works best: Having an assistant directly inside Discord keeps bench work collaborative and removes context-switching.
Platform constraints require smart prompts: Enforcing readable Unicode formatting at the prompt level solved Discord's lack of LaTeX rendering cleanly and effectively.
Specialized context beats raw model size: You do not need massive, slow models for domain-specific tasks. A targeted system prompt paired with a fast, lightweight model delivers an instant, practical workbench tool.
If you are participating in the Flip the Switch Hackathon or looking to automate your own team workflows, give Switch a try. It makes bringing AI agents into everyday team discussions surprisingly simple.
