VarAC AI Bridge
Bridging the Gap: My Dual-Band AI Gateway Project (HF/VHF EmComm Relay)
Introduction: I wanted to share a milestone in my shack that I believe has huge potential for the EmComm community. I’ve successfully engineered a Local AI Gateway that bridges HF and VHF bands, providing a "brain" for emergency communications that works even when the internet is completely dark.
The Problem: During a "Grid-Down" situation , we often have local VHF traffic that needs to jump to a long-distance HF net, or field operators who need technical answers (grounding, antenna lengths, medical protocols) without access to Google.
The Solution: The Gateway Using my primary rig (Ryzen 9 / RTX 3070 Ti) , I’ve built a three-tier system:
Tier 1: The Local Brain (Llama 3.2): I’m running a local instance of Llama 3.2 via Ollama. It’s fast, lightweight, and lives entirely on my hardware. No internet required.
Tier 2: The Multi-Band Bridge: I run dual instances of VarAC (HF and FM). By using a Symbolic Link to share a single VarAC.db database, any message or file received on local VHF/FM is instantly mirrored in the HF Outbox for long-distance relay.
Tier 3: The AI Assistant: I wrote a Python bridge that "listens" for radio queries. A station in the field can send a V-Mail asking for a technical solution, my local AI processes it in seconds, and beams the answer back automatically.
Recent Success: In recent testing, I queried the system for station grounding protocols. The AI processed the request and returned a concise, 38-word instruction—staying under the duty-cycle limit and providing vital info in a fraction of the time a manual voice exchange would take.
Why This Matters for ARES/RACES:
Zero-Infrastructure: It’s a self-contained "SDR-in-a-box" logic.
Cross-Band Interop: It allows a technician on a handheld to "talk" to the HF bands through the gateway.
Brevity & Clarity: The AI is tuned to be concise, keeping the frequency clear.
I’m hoping to peak interest in others wanting to adapt this. It turns our shacks from simple transceivers into intelligent communication hubs. I'm looking forward to testing and putting the polishing touches on this project.
73, KC5OLH (QRZ)
This is a great project! I suggest you post it also to the wider forum where we have much more audiance. This might be interesting for all. I am curios on how you are listening to incpming DATA (scanning the VarAC.db file?) and how to generate an answer (inset into the vmail table in the VarAC.db?) If it is - Cleaver :) if other way - also interested to hear :) 73s!