The counterintuitive method to get 10x better LLM outputs.
To my fellow clinicians and builders:
I want to give you the most important piece of advice I can offer about using AI in your practice, your codebase, and your life.
Right now, almost everyone is making a fatal mistake when they interact with Large Language Models. They spend hours trying to learn complex "prompt engineering" frameworks, downloading prompt libraries, and tweaking keywords like they are casting a magic spell.
But none of that solves the real bottleneck: Low-fidelity input.
When you use the default input method—the keyboard—you are instinctively filtering your thoughts. You compress the patient’s history, strip away the nuance of the business problem, or simplify the architectural logic of the code you're trying to write because the act of typing is inherently slow and tedious.
So, what is the #1 AI Prompting Secret to fix this?
When you talk, you naturally provide richer, deeper context. You explain the why, not just the what. You capture the messy nuances. When you speak to an AI, you aren't just giving it a command; you are giving it the entire landscape of your reasoning.
This applies to writing emails, drafting business strategies, brainstorming application features, or summarizing a clinical encounter. Voice is the ultimate leverage because it allows you to dump context at the speed of thought.
Here is exactly how I capture complex ideas, dictate code architecture, and structure documents using this secret:
I press a single hotkey to activate my local voice tool. I don't try to sound smart. I don't structure my sentences. I just brain-dump the messy, unfiltered reality of the problem I'm trying to solve or the idea I want to build.
I release the hotkey. The local voice model transcribes the messy audio instantly. The local LLM takes that context-rich transcript and runs it through a specific prompt (e.g., "Extract the core architectural constraints and draft an actionable plan").
The perfectly structured, formatted text is instantly pasted into my IDE or notes app.
Sixty seconds. Zero typing. 10x the context.
There are many dictation tools on the market, from built-in OS features to big tech cloud solutions. You can definitely start your voice journey with any of them.
However, as a physician, I needed something different. I couldn't send patient data to a public cloud, and I needed an AI reasoning engine that understood clinical nuance without compromising speed. I built Murmur to solve this exact problem.
Designed by a doctor for doctors, Murmur takes this 60-second capture method and runs it entirely locally on your hardware. It captures your thoughts and structures them instantly. Because no data is sent to a corporate cloud, you can safely talk out loud about sensitive business strategy, proprietary codebase logic, or even PHI.
The keyboard is legacy technology. Your voice is the ultimate interface.
Because you are reading this guide, I want to make it as frictionless as possible for you to try a privacy-first, local workflow.
Learn more and download at:
[Insert Murmur URL Here]