This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle
What I Built
I built LocalDocuShield 🛡️ — a privacy-first AI assistant for solo medical/legal/accounting professionals who handle sensitive client documents.
It helps users:
- summarize intake notes and long documents,
- extract action items,
- flag urgent risk levels (Low/Medium/High),
- and search prior case summaries semantically.
Who would pay for this?
- Solo doctors, small clinics
- Independent lawyers / paralegals
- Freelance accountants / tax preparers
What problem does it solve?
These users want AI productivity, but they cannot safely upload confidential records to cloud APIs without expensive compliance contracts.
LocalDocuShield runs AI inference on-device with open models, so data stays local.
Demo
- GitHub Repo: https://github.com/AslamSujah/LocalDocuShield
-
Local Demo: Run with Streamlit (
streamlit run app.py) and test with sample medical/legal text.
Code
The full source code is available here:
👉 https://github.com/AslamSujah/LocalDocuShield
Core files:
-
app.py– Streamlit app + local AI pipeline -
requirements.txt– dependencies -
README.md– setup/run guide
How I Built It
Stack
- UI: Streamlit
- Local LLM inference: Ollama + Gemma (open-weight model)
- Embeddings + semantic retrieval: local embeddings + SQLite storage
- Tabular anomaly scan: TabPFN
-
Persistence: SQLite (
localdocu_vault.db)
Pipeline
- User pastes text or uploads PDF/TXT.
- App sends content to local Ollama runtime (
localhost:11434). - Gemma returns structured JSON:
- client/case ID
- executive summary
- action items
- risk flag
- Summary + metadata are stored locally.
- User can run semantic search over prior summaries.
- Optional CSV ledger scan uses TabPFN to detect suspicious rows.
Why Does Open Innovation Matter?
Open innovation made this project viable in a way closed APIs often can’t for this market:
- Privacy by architecture: open-weight models run locally.
- Cost control: no per-token API billing for every client document.
- Accessibility: small firms can use AI without enterprise contracts.
- Customizability: workflows can be adapted per practice area.
Without open-source/open-weight AI, this kind of budget-friendly, privacy-preserving side-hustle product would be much harder to launch.
My Agent Session
I used an AI coding assistant workflow to iterate on architecture, deployment options, and implementation details while keeping the runtime local-first.
My First User
I validated the concept with small-practice-style workflows (medical/legal intake examples) and designed pricing/positioning around solo professionals.
Planned next validation step:
- onboard 1–3 real pilot users,
- measure time saved per intake,
- and test willingness to pay in the $19–$49/month range.
Prize Categories
I’m entering these partner-aligned categories:
- Best Use of Gemma (local summarization + structured extraction)
- Best Use of TabPFN (tabular anomaly/risk scan)
- Side Hustle Theme (clear business model + target buyer)
Thanks for reading! Feedback, feature requests, and collaboration ideas are welcome.
Top comments (0)