Apppublic

josephrw/doctor-verify-institutional

sourceHugging Faceupdated 4mo agoView on Hugging Face

0likes

App README

Doctor Address Verifier — Institutional Edition

Hugging Face Spaces deployment of a production-grade, multi-source doctor address verification system.

Features

—10 Verification Sources: NPI Registry, CMS Physician Compare, Google Places, Bing, Yelp, RateMDs, Healthgrades, SAMHSA, NY OPD, LLM Web Crawl
—LLM-Powered Web Crawl: Ollama analyzes actual web pages via DuckDuckGo search (falls back to keyword matching if unavailable)
—Parallel Source Fetching: ThreadPoolExecutor runs sources concurrently
—Pydantic v2 Models: Strict validation at all boundaries
—Immutable Audit Logging: SHA-256 checksums on every event
—SQLite Cache: 24hr TTL for scraped results
—Phone Extraction: Automatically extracts US phone numbers from all sources
—Consensus Engine: Weighted scoring by source importance
—Streamlit UI: Single verify, batch verify (Excel), system status, audit log
—Zero Configuration: Works out of the box with no API keys or environment variables required

Deployment

1. Create HF Space

bash

huggingface-cli repo create doctor-verify-institutional --type space --sdk docker

2. Push Code

bash

git clone https://huggingface.co/spaces/YOUR_USERNAME/doctor-verify-institutional
cd doctor-verify-institutional
# Copy all files from hf_institutional/
cp /path/to/hf_institutional/* .
git add .
git commit -m "Initial institutional deployment"
git push

3. Configure Secrets (Optional)

The system works without any secrets. For enhanced functionality, you may add:

Secret	Description	Effect
`GOOGLE_API_KEY`	Google Places API key	Enables Google Places source
`BING_SEARCH_KEY`	Bing Web Search API key	Enables Bing search source
`OLLAMA_HOST`	Ollama endpoint URL	Enables LLM-powered web crawl
`OLLAMA_MODEL`	Ollama model name	Uses specified model (default: llama3.2)
`APP_PASSWORD`	Access password for the UI	Adds password protection
`MAX_BATCH_SIZE`	Max records per batch	Adjusts batch limit (default: 100)

Default behavior without secrets:

—NPI Registry, CMS Compare, Yelp, RateMDs, Healthgrades, SAMHSA, NY OPD: Fully functional
—Web crawl: Uses keyword matching (no LLM required)
—Google Places, Bing: Disabled (optional enhancements)
—Authentication: Disabled (public access)

Architecture

app_hf.py      → Gradio UI (tabs: Single, Batch, Status, Audit)
engine.py      → Consensus engine + batch orchestrator
sources.py     → 9 independent source runners (each retry-wrapped)
llm_crawler.py → DuckDuckGo search + Ollama page analysis + cache
models.py      → Pydantic v2 models for all data boundaries
config.py      → Environment-backed config with security constants
audit.py       → Immutable JSONL audit log with SHA-256 checksums

Security

—Input Validation: All inputs validated via Pydantic models
—Audit Trail: Every verification logged with checksums
—Privacy: IP addresses hashed (SHA-256) before logging
—Rate Limiting: Configurable per-minute rate limits (default: 30/min)
—Least Privilege: No chmod 777, no wide-open CORS

License

Proprietary — Internal use only.