MeetCap #3: Core Discord Bot Features | Voice Recording, Whisper.cpp, VAD & Local LLM
In this third MeetCap devlog, I walk through the main implemented feature slices of my Discord meeting bot project.
it's okay, we'll get through it, I've been in this situation 100 times
when have you ever been in this situation?!
never, I just wanted to cheer you up
This is not a completed MVP yet. Some endpoints and read-side commands are still not fully implemented, and this video does not cover VPS deployment, Ansible structure, or production infrastructure. The focus here is the local development environment and the core features already built for the MeetCap solution.
In this video, I review the implementation history starting from shared contracts and moving through Discord bot commands, storage, voice recording, raw and processed audio artifacts, worker API integration, whisper.cpp, VAD, speaker-aware handling, transcript correction, and the local LLM summary pipeline.
Chapters:
00:00 MeetCap #03: core features, not full MVP
04:23 How the bot, worker, and local AI fit together
10:22 Discord bot commands and meeting flow
22:10 Voice recording: RTP packets, raw files, and artifacts
30:04 Teaching Whisper technical vocabulary
47:35 Unexpected guest: “People do not sleep, spies everywhere”
51:47 VAD, silence gaps, and Whisper hallucinations
01:05:01 Local LLM summary pipeline
01:33:53 BlackoutPlug with Mykola: first real flow meeting
01:39:05 Final transcript and generated summary review
Topics covered:
Project goal and local architecture
Bot / worker separation
Shared domain, storage, and worker contracts
Discord bot commands and local storage
Discord voice recording PoC
Raw and processed audio artifacts
whisper.cpp process boundary
VAD and silence filtering problems
Speaker-aware audio and SSRC-based handling
Local LLM summary architecture
Summary profiles and JSON/Markdown output issues
Local MVP test run and limitations
Tech stack and concepts:
Discord Bot, Rust, Songbird, whisper.cpp, whisper-cli, local LLM, llama.cpp server, GGUF models, Docker Compose, speech-to-text, VAD, speaker-aware transcription, local AI, meeting summarization, MVP architecture.
This is part of the MeetCap project: a self-hosted Discord bot for recording team meetings, transcribing audio locally, and generating structured summaries without sending everything into external AI tools.
GitHub demo repository: https://github.com/Korop/meet-cap-public
#MeetCap #DiscordBot #WhisperCpp #LocalLLM #Gemma4 #vps #codex #rustlang #voicerecognition #selfhosted