Back to the portfolio

Speech intelligence

Kotib AI

From recorded meetings to transcripts and structured minutes.

Created with my team

At a glance

Kotib AI addresses the work of converting meeting recordings into readable transcripts and structured minutes. Its intended users include meeting organizers, ministry leadership, and staff preparing the official record. Our team combines audio processing, speaker diarization, speech transcription, and AI-assisted document preparation, with editing and retranscription workflows. The documented engineering work includes evaluating Pyannote and NeMo against reference recordings and considering recording, background uploads, and meeting-platform integration together. These capabilities support preparation of a meeting record. Public materials do not establish an accuracy benchmark or the extent of deployment.

Kotib AI project presentation cover in Uzbek
Project presentation · original language: Uzbek

Created with my team

The projects in this portfolio are a shared effort. Product decisions, architecture, engineering, and delivery belong to the people who build together. My responsibilities vary from project to project.

01

The problem

Turning meeting audio into an accurate record requires separating speakers, transcribing speech, and organizing the discussion into a useful document.

02

Who it serves

Meeting organizers, ministry leadership, and staff preparing minutes.

03

What we build

Our team combines audio processing, speaker diarization, transcription, and AI-assisted minutes. Users can edit speaker names, share results, and request retranscription.

04

Engineering decisions

Diarization work evaluates Pyannote and NeMo against reference recordings. Recording lifecycle, background uploads, and meeting-platform integration are treated as parts of the complete product.

05

Capabilities

Speaker separation, noise processing, editable transcripts, automated minutes, sharing, and desktop/mobile workflows.

06

Context & limitations

Recording conditions, overlapping speech, and speaker separation affect quality. Generated transcripts and minutes need human review; no accuracy percentage is claimed here.

Expertise

Technical foundation

PyannoteNVIDIA NeMoSTTAudio processing
Explore the portfolioNext projectKIQTT