MEETING INTELLIGENCE · HEBREW-FIRST

MeetSum

Every meeting becomes searchable memory, tasks and automation — in Hebrew and English.

Talk about MeetSum meetsum.realization.co.il

IN SHORT

MeetSum is a self-hosted meeting-intelligence platform that turns Google Meet, Drive, desktop and uploaded recordings into transcripts, decisions and tasks. It handles Hebrew and mixed Hebrew–English speech properly and pushes results into Google Workspace, n8n, RealizeOS and AI agents.

interface languages, including RTL Hebrew: 5

CASE STUDY

Problem. Approach. Outcome.

The problem: Our meetings run in Hebrew and English, often in the same sentence. We needed transcripts that hold up, on our own servers, that reach the systems where the work happens.
The approach: One retry-safe job pipeline for every source. Speech is routed by language and length: a Hebrew-tuned local model for short meetings, a cloud model for long or mixed ones, with every stage’s model and confidence recorded.
The outcome: Live in production with import, processing, review, sharing, export and automation working end to end. Cost scales with usage, not seats.

Case in point: making local Hebrew speech recognition safe

A 52-minute mixed Hebrew–English recording crashed the local speech container in production. The pipeline fell back to the cloud model automatically, the meeting still completed, and users saw a quality warning rather than a silent failure. The fix was a routing policy: local recognition for Hebrew under fifteen minutes, cloud for longer or mixed audio — plus chunking and memory caps for the next release.

IMAGES

The work, shown.

MeetSum meeting workspace with a meeting list, processing pipeline and an AI summary of an acquisition review — Meeting workspace: summary, decisions, tasks and the processing pipeline · demo data

The same MeetSum workspace in the Hebrew right-to-left interface — The same workspace in Hebrew, right to left · demo data

Source. MeetSum project overview v0.6.0, September 2026.

RELATED FIELD NOTES

The thinking behind it.

AI in practice

Making Hebrew speech recognition production-safe

A 52-minute mixed-language meeting crashed our local model. The fix was a routing policy, not a bigger server.

NEXT STEP

Meetings that should feed your systems?

Or book a 20-minute intro call with Asaf.

Talk about MeetSum