Trust But Verify

Trust But Verify combines Databricks AI, OpenAI web verification and human oversight to turn healthcare facility claims into evidence-backed planning decisions.

Comment

Inspiration

Healthcare planners often rely on facility datasets containing thousands of records with inconsistent descriptions, incomplete information, and unverified capability claims. We wanted to answer a simple but important question: Can a healthcare facility actually do what it claims?

What it does

Trust But Verify evaluates healthcare facility capabilities such as ICU, NICU, emergency, trauma, maternity, and oncology services. It combines internal trust scoring from Delta Share healthcare data with external verification using OpenAI-powered public web evidence to generate explainable trust signals, citations, confidence scores, and planner recommendations.

How we built it

We used Databricks to ingest and analyze healthcare facility data shared through Delta Share. AI-powered extraction and trust scoring identify evidence from facility records, while OpenAI validates claims using public web sources. A Trust Reconciliation Engine compares internal and external evidence and produces actionable recommendations for planners.

Challenges we ran into

Healthcare data was highly inconsistent, with varying terminology, incomplete records, and uneven levels of detail. Public information quality also varied significantly across facilities. Balancing trust, uncertainty, explainability, and human oversight was one of the most challenging aspects of the project.

Accomplishments that we're proud of

We successfully built a solution that does not blindly trust either the dataset or the internet. By combining Databricks intelligence, OpenAI verification, and human judgment, we created an explainable decision-support system that highlights verified facilities, data gaps, suspicious claims, and areas requiring further review.

What we learned

We learned that trust is not binary. Datasets can be incomplete, web information can be outdated, and AI models can be overly confident. The most valuable insight was that decision-makers need transparent evidence and clear reasoning rather than a simple yes-or-no answer.

What's next for Trust But Verify

We plan to expand support for additional healthcare capabilities, incorporate more trusted healthcare data sources, improve confidence scoring, add geographic planning insights, and enable continuous verification as facility information changes over time. Our long-term vision is to create a trusted healthcare intelligence platform for planners, NGOs, and public health organizations.

Built With

  • acid-transactions
  • ai
  • ai-powered
  • amazon-web-services
  • api
  • apis
  • bm25-keyword-search
  • catalog
  • css
  • data
  • databricks
  • databricks-apps
  • databricks-sql-connector
  • databricks-sql-warehouse
  • databricks-vector-search
  • databricks-vectorsearch
  • delta
  • delta-lake
  • engineering
  • evidence
  • extraction
  • genie
  • github
  • human-in-the-loop
  • hybrid
  • information
  • lake
  • language
  • large
  • models
  • natural-language-processing
  • next.js
  • openai
  • pandas
  • pipelines
  • pyspark
  • python-3.11
  • react
  • reconciliation
  • responsible
  • rest
  • scoring
  • search
  • semantic-vector-embeddings
  • serverless-compute
  • sharing
  • sql
  • sqlite
  • streamlit
  • tailwind
  • trust
  • typescript
  • unity
  • unity-catalog
  • vector
  • web
Share this project:

Updates

Submission history