Digisensus

is Conversational Intelligence system that converts your business agent-client conversations into action to improve your business!

[Digisensus @Digisensus]

Digisensus is not another analytics dashboard.

It scores 100% of your calls, chats and emails against your own scorecard, then reads thousands of conversations in one deep-dive audit to find what single reviews never see.

Free 30-min demo → https://t.co/ZJRsFcPqkv

[Digisensus @Digisensus]

@rflynn_ryan Good list. I'd add a fourth: does a bot need to join the call at all? Recording on your own Mac keeps the file on that machine instead of a vendor's cloud, and consent is still on you. We built Digisensus Recorder for that, free and open source: https://t.co/Z6bjpeiJyq

[Digisensus @Digisensus]

@KNash56325 Short, flat answers and the caller repeating themselves are the usual tells, plus anyone saying "person" or "agent" twice. Hang-ups right after the bot repeats a question are another. Track those and the owner learns fast which calls to keep human.

[Digisensus @Digisensus]

@KenAdamsGB Putting AI and human agents on one scorecard is the right direction. One thing worth asking for when contact centers set it up: a quote from the call behind every score. That's what lets a QA lead trust or challenge a grade, especially on AI calls.

[Digisensus @Digisensus]

@Bic_Revelation The idempotency point is spot on, double charges are exactly what green test suites miss. The other half is after release: real customers find failures no fault injector will, like a clean API call paired with a promise the agent shouldn't have made.

[Digisensus @Digisensus]

@TenthRevGroup Honest answer for most teams: when a customer complains. Faster is checking what the agent actually told customers, every conversation, against the company's own policy, with the quote kept. Ideally that check sits outside the team that built the agent.

[Digisensus @Digisensus]

@Rajesh7113 Your third point is the one that matters most. Weekly scoring works best when it covers every chat, not a sample, and uses the same rubric as the human team, so you can see where the AI is quietly worse. That's what Digisensus does: https://t.co/VrmStDbnLk

[Digisensus @Digisensus]

@MakadiaHarsh Great list, and the last line is the hardest to check. After launch, pull every call where the caller asked for a person or went quiet, and see if the agent actually handed off. Those calls rarely show up as errors. We score every AI call for that at https://t.co/VrmStDbnLk

[Digisensus @Digisensus]

@EternalBahamut @NVIDIAAI Whisper only gives you text, so who said what needs a separate diarization step. This model does speaker labels and translation in one pass. NVIDIA's release targets their GPUs, but an MLX port for Apple silicon was announced yesterday, so running it on a Mac looks doable.

[Digisensus @Digisensus]

@sharyph_ Digisensus Recorder: a Mac app that records your Zoom, Meet or Teams calls without sending a bot in, keeps the audio on your machine, and writes notes with whichever AI you like, or none at all. Free and open source. https://t.co/Z6bjpejhnY

[Digisensus @Digisensus]

@ShouldiceJason Curious who checks Sarah. If an AI is coaching the floor, I'd score a sample of the calls she flags against the same QA rubric your human leads use, with quotes from the call as evidence. That's how you tell good coaching from confident coaching.

[Digisensus @Digisensus]

@Qorne Nice setup. Two things that paid off for us on client calls: run a voice activity pass first so Whisper doesn't invent text over long silences, and if you can capture each side on its own track, speaker labels come for free instead of from diarization.

[Digisensus @Digisensus]

@MelissaBaker88 The model is guessing ownership from who spoke last. Two fixes that help: say the owner out loud before moving on ("Sam has this, due Friday"), and tell the AI to assign an item only when someone clearly agreed to it, otherwise list it as unowned.

[Digisensus @Digisensus]

@notathrv Raycast for launching apps and clipboard history, Rectangle for snapping windows, and AltTab if you're coming from Windows. Then put your settings in iCloud or a dotfiles repo early, so the next Mac takes an hour to set up instead of a weekend.

[Digisensus @Digisensus]

@robjama Fun idea. Logging bpm per commit instead of per PR would be interesting too, you'd see whether the stress is in writing the fix or in waiting for CI. Does the watch connection hold up when the Mac sleeps?

[Digisensus @Digisensus]

@kualiplli Same. If you're the one who wants notes, you can record on your own Mac instead, so nobody else gets a bot in their call. We built Digisensus Recorder for exactly that, free and open source: https://t.co/Z6bjpeiJyq

[Digisensus @Digisensus]

@Zayyan_Dalai @bot Rotate it. A different person takes notes each meeting, and the template is just decisions, owners and dates. If nobody wants it, record the call, let AI draft the recap, and the meeting owner sends it. Notes shouldn't default to whoever cares most.

[Digisensus @Digisensus]

@BLSmith2112 @stevenmarkryan @bot Nice pipeline. One tweak that helped us: have Whisper output timestamps and keep the raw transcript next to the Ollama summary, so when a summary line looks off you can jump to the exact moment instead of trusting the model.

[Digisensus @Digisensus]

@thisisharding Speaker labels get a lot more reliable when each side is recorded on its own track, so the model never has to guess who's talking. For remote interviews on a Mac we built Digisensus Recorder for that, free and open source: https://t.co/Z6bjpeiJyq

[Digisensus @Digisensus]

@rwessman If you keep it on, give the recap 30 seconds before it goes out. Auto summaries nail the topics and often miss who actually agreed to do what, and that's the part people act on.

[Digisensus @Digisensus]

@ramanvibes @wespreadjam Plenty try. The weekend build usually stalls on capture, not the summary: pulling the other side's audio on macOS without a bot and keeping your mic apart. We open-sourced that part as Digisensus Recorder if anyone wants a head start. https://t.co/Jiaa8tL6VU