Live Call Coaching Desktop App
Budget: £250 – £750 GBP
I handle a lot of live phone conversations and need a lightweight coaching aid that keeps up with me, not the other way around. The idea is simple:
• While my phone is on speaker, the app constantly captures microphone input.
• Audio is pushed through Whisper (or another low-latency speech-to-text pipeline) so I see a near-instant transcript.
• Every three to five seconds the most recent transcript chunk is sent to the OpenAI Chat Completion endpoint.
• The response comes back as one or two concise suggestions—each under twelve words—about what I could say next.
• On screen I want two clean panes: running transcript on the left, fresh suggestions on the right.
Crucial performance details
• End-to-end delay, mic to on-screen suggestion, must stay below three seconds.
• No chatbot or automated voice agent—this is purely a human-in-the-loop coaching tool.
Tech decisions already made
• macOS target.
• Python backend with a simple, functional PyQt interface.
Delivery expectations
1. A bundled macOS app I can launch with a double-click.
2. Full source code with clear setup notes so I can drop in my own OpenAI key and tweak timing parameters.
3. Brief README and, ideally, a short screen-capture demo proving the latency target.
I’ll test early builds on real calls and give fast feedback, so iterative releases are welcome. If you have ideas to shave milliseconds or keep the UI ultra-clean, let’s build them in.
• While my phone is on speaker, the app constantly captures microphone input.
• Audio is pushed through Whisper (or another low-latency speech-to-text pipeline) so I see a near-instant transcript.
• Every three to five seconds the most recent transcript chunk is sent to the OpenAI Chat Completion endpoint.
• The response comes back as one or two concise suggestions—each under twelve words—about what I could say next.
• On screen I want two clean panes: running transcript on the left, fresh suggestions on the right.
Crucial performance details
• End-to-end delay, mic to on-screen suggestion, must stay below three seconds.
• No chatbot or automated voice agent—this is purely a human-in-the-loop coaching tool.
Tech decisions already made
• macOS target.
• Python backend with a simple, functional PyQt interface.
Delivery expectations
1. A bundled macOS app I can launch with a double-click.
2. Full source code with clear setup notes so I can drop in my own OpenAI key and tweak timing parameters.
3. Brief README and, ideally, a short screen-capture demo proving the latency target.
I’ll test early builds on real calls and give fast feedback, so iterative releases are welcome. If you have ideas to shave milliseconds or keep the UI ultra-clean, let’s build them in.