I Stopped Saying "Let Me Get Back to You"

Last Tuesday, someone asked "What's the deployment timeline?" in our sprint planning. Before I could open my mouth to say "let me find that doc," the answer was already on screen.

From the tech spec. Page 12. Exact quote. With a link.

Nobody searched. Nobody interrupted the flow. The meeting just... continued.

The Problem Everyone Has (But Nobody Talks About)

You're in a meeting. Someone asks a question. You have three terrible options:

1. Say "let me get back to you" → Question gets forgotten. You look unprepared.
2. Search while everyone waits → 3 minutes of silence. Meeting stops. Everyone stares.
3. Answer from memory → 50/50 chance you're wrong. Decisions made on bad info.

All three options suck. So meetings become longer, less productive, and nobody gets clear answers.

Last month, I counted: In a typical 1-hour product meeting, we said "let me get back to you" 7 times.

Seven questions. Seven follow-ups I needed to remember. Seven topics we'd revisit next meeting.

That's not a meeting. That's a waste of everyone's time.

What We Did Differently

Instead of waiting for the meeting to end and searching later, we built AI that watches the conversation and searches in real-time.

When someone asks "What's the deployment date?" here's what happens:

Step 1: AI detects the question

As the words are being spoken. Not after. During.

Step 2: Four places to look

📚 Your uploaded documents - Searches every doc, transcript, and note. Takes 2 seconds.

💬 Earlier in this meeting - Maybe someone already answered it 10 minutes ago. You just forgot.

👂 The next 15 seconds - Maybe someone's about to answer naturally. AI waits and listens.

🤖 AI's best guess - Only if nothing else works. Comes with a big warning: "This is AI-generated, verify it."

Step 3: Answer appears on screen

With the source. "From Q3 Budget Planning.pdf, page 12." Click to verify.

Total time: Usually under 2 seconds. Always with the source shown.

But Wait, There's More: Automatic Action Tracking

While monitoring for questions, the AI also detects action items and commitments:

As the conversation progresses, action items update with more details. No manual note-taking required.

The Technical Architecture (For the Curious)

Real-Time Processing Pipeline

1. Audio Capture: Browser captures microphone audio (PCM 16kHz, 16-bit, mono)
2. WebSocket Streaming: Audio chunks sent to backend every 100-200ms
3. AssemblyAI Transcription: Real-time speech-to-text with speaker diarization
4. GPT-5-mini Streaming: Analyzes transcript as it arrives, detects questions and actions
5. Answer Discovery: Four parallel tiers run simultaneously
6. Client Updates: Progressive results streamed back to UI

Cost Optimization

We obsessed over cost because enterprise meeting tools are ridiculously expensive:

• AssemblyAI: $0.90/hour (real-time transcription + speaker diarization)
• GPT-5-mini: ~$0.15/hour (streaming intelligence)
• Total: $1.05/hour per meeting

For comparison, Otter.ai Business costs $20/user/month. If you have 10 users and 20 meetings/month, that's $200/month = $10 per meeting hour.

We're 10x cheaper.

The Experience: What Using It Feels Like

Here's what happened in a real product planning meeting last week:

No one had to stop to search. No one had to take notes. The meeting flowed naturally, and all the information was captured automatically.

100% Source Transparency

Every answer is labeled with its source:

• 📚 From Documents - Searched your knowledge base
• 💬 Earlier in Meeting - Found in current meeting transcript
• 👂 Answered Live - Someone responded naturally in conversation
• 🤖 AI Answer - GPT-generated with disclaimer (rare)

You always know if you're getting information from your own documents or if the AI is guessing. No black box.

Why This Matters

The average knowledge worker spends 2.5 hours per day searching for information (IDC Research). That's 31% of the workday.

In meetings, this manifests as:

• "Let me get back to you" (and then forgetting)
• Long pauses while someone searches
• Same questions asked repeatedly
• Decisions delayed because information isn't available

Our real-time intelligence system eliminates this friction. Questions get answered immediately. Decisions move forward. Meetings are shorter.

Try It Yourself

TellMeMo is 100% open source and free to self-host. Or use our cloud version to get started immediately.

Set up takes 5 minutes. You'll need:

• OpenAI API key (for GPT-5-mini streaming intelligence)
• AssemblyAI API key (for real-time transcription)
• Some documents to create your knowledge base (optional but recommended)

Start your next meeting with the AI Assistant toggle ON. Watch as questions get answered automatically.

What's Next?

We're already working on v2:

• Multi-language support - Currently English-only, expanding to 50+ languages
• Custom confidence thresholds - You control when AI fallback triggers
• Integration with Zoom/Meet - Direct plugin for video conferencing platforms
• Smart follow-ups - AI suggests follow-up questions based on incomplete answers
• Answer quality scoring - Confidence levels for RAG search results

If you want to influence the roadmap, join our GitHub community.

The Future of Meetings

We believe the future of meetings isn't about recording everything. It's about making information instantly accessible when you need it.

No more "let me get back to you."
No more searching during meetings.
No more forgotten questions.

Just answers. In real-time. From your own knowledge base. With complete transparency.

That's what we built.

---