Your AI meeting notes app works. But transcription costs are eating your margins. Accuracy isn’t up to par. Audio capture is harder than it should be (especially with Electron). Users don’t want their private conversations sent to 3rd parties. Permissions are confusing (and annoying your users). Diarization seems like a pipe dream.

I fix these problems.

Problems I solve

Performance tuning

Meeting transcription apps are a dime-a-dozen. Your niche is how you handle the trade-offs between speed, cost, and accuracy.

I find the bottlenecks and fix them:

  • Cut costs through better preprocessing and provider selection
  • Improve speed with smarter batching and chunking
  • Boost accuracy with the right model for your use case
  • Balance all three for your specific constraints

On-device strategy

Privacy, speed, offline use, and latency can all be improved by sidestepping the cloud. On-device pipelines now support speech-recognition, voice activity detection, diarization, and keyword detection.

I help you figure out what belongs on-device:

  • Speed vs accuracy
  • Battery vs performance
  • Hybrid approaches (both on-device and cloud)

Frontier experimentation

Want to see how the latest model works with your app? Before you commit to a major architectural shift, we validate it. This is a new frontier. But the playbook is becoming clearer.

  • Side-by-side model benchmarks
  • Build proof-of-concepts for new features
  • Experiment with new transcription techniques
  • Audio processing experiments
  • Real-world meeting dataset evaluation

Electron audio capture

Reliable audio pipelines are harder than they look — especially in Electron.

Capturing audio is only half the problem. Moving it efficiently between processes and into your model without latency is where most systems break. Especially on lower-cost devices. In Electron apps, audio capture, inter-process communication (IPC), and model placement directly determine performance and reliability.

Common problems I can solve:

  • Where should the model live, and how should raw audio move through the system?
  • Performance bottlenecks getting the audio to your model
  • Permissions headaches
  • Ambiguous error messages that don’t tell you what’s actually wrong
  • Audio capture that fails on certain devices for no clear reason
  • Preventing feedback loops and echo

Why me

Native, systems, and audio expertise.

Most devs working on AI meeting notes apps come from web backgrounds. React, TypeScript etc. They know frontend. But transcription performance problems aren’t frontend problems. They’re systems problems. Audio capture, IPC, memory management, battery life, on-device ML. This is native platform work.

When your Electron app is dropping audio frames on certain devices, you don’t need another React developer. You need someone who understands audio buffers and IPC.

I’ve been doing client and systems programming for decades. With audio experience going back to my university days.

How this works

Here are ways we can work together:

Roadmapping

You tell me what’s holding your app back. I give you a plan of attack. Typical engagement: 2 days - 1 week.

Short-term improvements

You’re not ready for a full overhaul. Let’s find the next improvement that moves the needle. Usually 1-4 weeks of hands-on work. I ship production fixes and unblock your team.

Long-term retainer

You need ongoing performance work. Monthly engagement where I’m available for optimization work, architecture decisions, and firefighting. Good fit if you’re scaling and need consistent performance expertise.

Full-time

An option for the right opportunity. IC or Leadership. Let’s do this!

Let’s talk

If your AI meeting notes app has performance problems, transcription cost issues, or accuracy challenges, email me at jaim@sharpfivesoftware.com.

Include:

  • What problem you’re trying to solve
  • Your current tech stack (Electron? Native macOS? Which transcription provider?)
  • Your timeline