Amateur radio. AllStarLink. Local AI.

KK7NQN AI Net Logger

A self-contained radio analysis stack for AllStarLink hub nodes. The system records node audio directly from Asterisk, transcribes traffic locally, detects callsigns, verifies operators through QRZ, and groups conversations into searchable net logs.

Diagram of AllStarLink hub nodes flowing into transcription, callsign detection, QRZ verification, and net analysis.

Project overview

Built for the kind of radio traffic that does not fit in a spreadsheet.

Direct AllStarLink capture

Audio is recorded from AllStarLink hub activity through Asterisk, with no physical sound card input or speaker loopback required.

Local transcription

The main Z840 server handles hub coordination, storage, and Whisper transcription using its RTX 3060 GPU.

Heavier AI on Sable

Sable runs the deeper analysis jobs, including callsign extraction, wider net grouping, and large-language-model passes.

Operator friendly logs

Detected callsigns, transcript segments, receive state, connections, and analysis passes are tracked for review and later search.

Analysis pipeline

Every transcript gets an analysis trail.

The project keeps a verbose record of each analysis step so operators can see how a transcript moved through the system. Steps can be extended later as new modules are added.

  1. 01Hub audio chunked

    Growing AllStarLink recordings are split into processable chunks.

  2. 02Whisper transcription

    Speech becomes text, confidence, timing, and processing metadata.

  3. 03Callsign extraction

    Candidate callsigns are detected and checked before being stored.

  4. 04QRZ verification

    Known operators can be enriched with public QRZ profile data.

  5. 05Net grouping

    Recent transcript context helps collect a whole net instead of isolated fragments.

  6. 06AI analysis pass

    Large-model analysis can summarize, classify, and prepare future reports.

Monitored hub targets

Current AllStarLink focus

66296

PSRG

Hub node connected toward PSRG activity for recording and net detection.

683211

DoDropIn

Hub node monitored with duplicate-aware grouping when traffic crosses linked systems.

683210

FireSide

Hub node tracked alongside related nets so repeated linked audio can be recognized.

Public status

Running as an operator-facing experiment.

The live dashboard is kept private on the local network while this public page documents the project for operators, search crawlers, and anyone curious about the build.

Stack
Python, FastAPI, PostgreSQL, RabbitMQ, Redis, Whisper, vLLM
Primary host
Z840 main server and transcription worker
AI host
Sable GPU worker for heavier analysis passes
Goal
Readable, searchable logs for amateur radio nets and linked-system activity