Palora — AI Meeting Notes Without the $20 Price Tag

Inspiration

Granola and Fireflies showed how powerful automated meeting notes can be. But they’re expensive often $20–$30/month and rely on bots joining your calls, sending invites, and clogging your calendar.

We wanted the same experience in a lighter, faster, cheaper, and more private app. No bots. No invites. No setup.

Just an app that automatically knows when you’re in a meeting and gives you clean minutes the moment you’re done.

That idea became Palora.

What it does

Palora detects when you're in a meeting by monitoring system audio + microphone audio (locally, on-device). It does not join your meetings as a bot.

After the meeting ends, Palora:

Records the conversation locally
Transcribes the audio
Summarizes the meeting into polished minutes
Extracts decisions and action items
Rewrites or restructures messy notes on command

It also transforms raw text bullet lists, brain dumps, rough ideas into clean, structured summaries with one tap.

How we built it

Native Swift app for speed, OS integration, and low overhead
Automatic meeting detection via real-time system + mic audio monitoring
Hybrid AI pipeline:
Local models (Whisper, LLaMA-based summarizers, Qwen) for on-device transcription & privacy
Cloud LLMs for high-quality summarization and structured output
Reusable prompt templates designed for consistency
Minimal, fluid UI that keeps interactions fast and intuitive
Production-ready architecture built to ship, not just demo

This combination allowed us to keep latency low, enhance privacy, and reduce API cost.

Challenges we ran into

Balancing local model speed vs. cloud model accuracy
Designing prompts flexible enough to handle chaotic real-world meetings
Making the app completely automatic—zero setup, zero friction
Ensuring audio detection doesn’t drain battery or miss events
Keeping the Swift codebase clean while building fast during a hackathon

Accomplishments we’re proud of

Shipping a fully functional, production-ready meeting notes app
Matching (and sometimes beating) the quality of tools like Granola
Implementing true automatic meeting detection without bots
Integrating local + cloud models in a seamless hybrid pipeline
Building something polished enough to launch as a paid product immediately

What we learned

Local models are fast and private, but require careful tuning
Prompt engineering matters more than people think—tiny changes = huge quality differences
Simple UX beats feature-heavy UX every time
Architecting like you’re shipping forces better design patterns and cleaner code