Architecture note
Available for software engineering roles - backend systems, production reliability, cloud deployment, and AI-assisted workflows.
A local-first, privacy-driven Digital Twin for Australian sole traders. No cloud. No subscriptions. No data leaving your machine.
01
Primary language: Python
Technologies: Python, Streamlit, ChromaDB, Pydantic, Pytest
Topics: Not specified
Last updated: 2026-04-11T01:14:42Z
Stars: 0
Forks: 0
Status: Active
Architecture note
Visual notes
The AIThatActually WorksFor You.
A local-first, privacy-driven Digital Twin for Australian sole traders. No cloud. No subscriptions. No data leaving your machine.
The Problem: Australian sole traders are drowning in admin. Invoicing, BAS, ATO compliance, email triage — it consumes 10+ hours every week. Existing "solutions" require cloud uploads, monthly subscriptions, and trust that big tech won't monetise your financial data.
The Solution: BackPocket OS runs entirely on your machine. Your business data never leaves your device. Using Agentic RAG and local large language models, it handles your business admin — generating invoices, classifying emails against Fair Work guidelines, preparing BAS-ready reports — while you stay compliant and in control.
Why Now:
The Market: 1.2M+ sole traders in Australia. $3.2B spent annually on business admin tools. We're building the alternative that puts owners first.
Python 3.12+
# macOS
brew install [email protected]
# Ubuntu/Debian
sudo apt update && sudo apt install python3.12 python3.12-venv
Ollama (Local AI Engine)
# macOS
brew install ollama
# Linux
curl -fsSL https://ollama.ai/install.sh | sh
Then pull your model:
ollama pull gemma4:4b # Recommended for speed
# or
ollama pull phi3.5 # If using Microsoft ecosystem
Whisper (Voice transcription)
ollama pull whisper
# Clone or extract this project
cd backpocket-os
# Create virtual environment
python3.12 -m venv venv
source venv/bin/activate # macOS/Linux
# or
venv\Scripts\Activate # Windows
# Install dependencies
pip install -r requirements.txt
# Start the app
streamlit run ui/app.py
That's it. Open http://localhost:8501 and start using your Digital Twin.
BackPocket OS is designed with defense in depth. Here's how we protect you:
| Layer | Protection |
|---|---|
| Storage | All data stays in ~/.backpocket/ — local SQLite + ChromaDB. Never sent to external servers. |
| AI Processing | Ollama runs locally. Prompt inputs and outputs never leave localhost. |
| Vector DB | ChromaDB stores embeddings locally. No cloud sync. |
| Network | App operates offline. Internet only needed for initial dependency install. |
| Threat | Mitigation |
|---|---|
| Data exfiltration | All I/O bound to local filesystem. No external API calls in core flow. |
| Model poisoning | Verify Ollama model checksums. Use GNUPG signatures where possible. |
| Memory scraping | Process isolation via OS user permissions. No shared memory exposed. |
| Supply chain | Dependencies pinned in requirements.txt. Audit with pip audit. |
This isn't just "privacy-friendly." It's privacy-native.
backpocket-os/
├── core/ # Business logic & models
│ ├── __init__.py
│ ├── invoice_engine.py # Invoice generation
│ ├── compliance.py # ATO/Fair Work rules
│ └── email_triage.py # Email classification
├── agents/ # AI Agent orchestration
│ ├── __init__.py
│ ├── orchestrator.py # Agentic RAG controller
│ └── prompts.py # System prompts
├── tools/ # Tool definitions
│ ├── __init__.py
│ ├── pdf_generator.py # PDF output
│ └── pdf_extractor.py # PDF ingestion
├── data/ # Data stores
│ ├── chromadb/ # Vector embeddings
│ │ └── index/
│ └── ingest/ # Source documents
├── ui/ # Streamlit frontend
│ └── app.py
├── config/ # Configuration
│ └── settings.py
├── TASKS.md # Non-tech contributor tasks
├── ARCHITECTURE.md # System diagram
└── requirements.txt
pytest tests/ -v
We need more than code. See TASKS.md for ways to contribute without writing a single line.
MIT License — because open source benefits everyone.
Built in Australia, for Australians.
Next field note
Open to software engineering roles across full-stack systems, platform and reliability work, workflow automation, and applied AI.
Write to Himanshu Lade →