Wasi AI App

A WasmCloud AI application that transcribes audio files using a ggml-tiny.bin model and summarizes the resulting text with a lightweight Qwen model, all orchestrated through a web UI. Built with Rust, WASI Components, and the WasmCloud. App uses a local LLM to generate meeting notes.

Components

1. Transcriber : Generate transcription for the Audio file
2. Summarizer : Summarize transcription
3. Web : UI

✨ 🚀 Start => 🎧 Transcriber => 📝 Summarizer => Done 🏁

Transcriber Component

• Generate transcription for the Audio file

• Audio file as Input: output.wav | Size: 141.3 MB

• AI Model to generate transcription: ggml-tiny.bin | Size: 77.7 MB

• Srack: Rust, oxiwhisper, Wasi, WasmCloud

Summarizer Component

• Summarize transcription

• Audio file transcription as Input: Text

• AI Model to Summarize transcription: Qwen3-0.6B-Q4_K_M.gguf | Size: 396.7 MB

• Srack: Rust, Candle, Wasi, WasmCloud

Web Component

• Web UI

• Workflow, Run, Status, Description

• Srack: Rust, JS, CSS, HTML, Wasi, WasmCloud

The files required in testdata dir

• testdata/Qwen3-0.6B-Q4_K_M/Qwen3-0.6B-Q4_K_M.gguf
• testdata/Qwen3-0.6B-Q4_K_M/tokenizer.json
• testdata/ggml-tiny.bin
• testdata/sample_audio.wav
• testdata/user-prompt.txt
• testdata/system-prompt.txt

Check the local k8s deployment guide

• deployment/readme.md
• deployment/deployment.yaml
• Deploy a Wasm Workload to Kubernetes