1
Private by default
Field notes, checklists and operational context should not be sent to external LLM APIs.
QVAC Hackathon I · Edge AI
Private field notes. Local answers.
Offline Field Copilot is a focused Android copilot that runs QVAC SDK inference on consumer smartphone hardware. Paste a checklist or field instruction, ask a practical question, and get an on-device answer — with no cloud LLM APIs for inference.
Mobile-first
On-device inference
Local context
Offline-capable after setup
QVAC runtimelocal
Field note loaded
Private checklist stays on the device.
Generator unit inspection: fuel level, oil leakage, cooling airflow, emergency stop, abnormal vibration, temperature warning.
First: verify safety-critical items.
Watch for: leakage, airflow obstruction, abnormal vibration, temperature warning.
Watch for: leakage, airflow obstruction, abnormal vibration, temperature warning.
No cloud LLM path
Designed for local QVAC inference. Demo evidence should show no OpenAI, Anthropic, Gemini or OpenRouter API configuration.
Product values
The product is not a generic chatbot. It is a small, verifiable edge-AI workflow for practical field work.
2
Offline resilience
The app is designed for low-connectivity situations and is offline-capable after model setup/cache.
3
Focused assistance
One local context, one practical question, one concise answer. No agent theater.
4
Verifiable build
Open-source repo, clear README, evidence folder, run logs, and no-cloud verification.
Workflow
A simple demo flow judges can understand in under 30 seconds.
01
Load model
QVAC SDK initializes the local model lifecycle on the Android device.
02
Add context
The user pastes a field note, checklist, instruction or local procedure.
03
Ask question
Example: “What should I do first and what risks should I watch for?”
04
Generate locally
A concise answer is generated on-device, then the model can be unloaded.
Why this matters
Field work often happens where cloud-first AI is inconvenient or inappropriate.
P
Private context
Operational notes may include sensitive work details. Keeping context on-device reduces unnecessary exposure.
R
Resilient execution
Low-connectivity environments need useful workflows that do not depend on every request reaching a remote LLM.
C
Cost control
Local inference avoids per-request cloud LLM API dependency for the demonstrated workflow.
Honest scope
The claim is intentionally narrow. That makes it stronger.
What it does
- Runs as a mobile-first Android app
- Uses QVAC SDK for local inference
- Works over pasted field notes and checklists
- Shows model/runtime/device status
- Uses no cloud LLM APIs for inference
What it does not claim
- Not production-ready
- Not enterprise-grade security
- Not a complete RAG, OCR or voice system
- Not guaranteed on every Android phone
- Not fully offline from zero setup
Evidence-first submission
The repo should prove the build, not just describe it.
Minimum evidence bundle
README
Setup and demo flow
Setup and demo flow
License
Apache-2.0
Apache-2.0
Device file
Phone, OS, RAM, chipset
Phone, OS, RAM, chipset
Run log
Load → generate → unload
Load → generate → unload
No-cloud note
No cloud LLM APIs
No cloud LLM APIs
Demo video
Physical phone in frame
Physical phone in frame
Tech & hardware
AppExpo / React Native / TypeScript
Inference@qvac/sdk
ContextLocal pasted field notes
DeviceRedMagic / Nubia Android phone
Cloud LLM APIsNot used for inference
Offline scopeAfter model setup/cache
Built for a clear judge reaction.
“This is a mobile QVAC app, it runs local inference on consumer Android hardware, and the evidence shows no cloud LLM dependency.”