Healthcare.gov Plan Q&A Demo

Asks your HCGov FastAPI (/v1/ask) and shows the answer.


        


🔍 What’s happening behind the scenes

This demo connects a custom Python FastAPI service to structured Healthcare.gov plan data and large language models.

When you click Ask, the system performs the following steps:


1) Plan data ingestion (offline)

Healthcare.gov plan documents are collected and processed ahead of time:

  • Summary of Benefits (SBC) PDFs
  • Drug formularies
  • Plan brochures

These documents are:

  • Parsed into structured JSON
  • Chunked into small, searchable excerpts
  • Stored in PostgreSQL with vector embeddings
  • Scored using a configurable weighted ranking system

This allows the system to reason over real plan documents — not summaries or guesses.


2) Plan ranking (deterministic)

Before any AI is involved, plans are ranked using a transparent scoring model.

The ranking system:

  • Penalizes high deductibles and out-of-pocket maximums
  • Penalizes missing or unclear benefits
  • Can be tuned for different health profiles (e.g. diabetes, cardiac risk, prenatal care)

This step ensures the AI only looks at the most relevant plans.


3) Retrieval (RAG mode)

When a question is asked:

  • The system searches the top-ranked plans
  • Relevant document excerpts are retrieved
  • Only those excerpts are passed to the language model

The AI cannot use outside knowledge — it answers strictly from cited plan text.


4) Answer generation with citations

The response you see is:

  • Generated from real plan excerpts
  • Grouped by plan ID
  • Cited down to the exact PDF and page number
  • Explicit about differences between plans

If the answer varies by plan, the system will say:

“It depends by plan.”

If the information is not present, it will say so.


5) Why this matters

Most plan comparison tools:

  • Hide their scoring logic
  • Don’t show citations
  • Can’t answer nuanced questions

This demo is built to show:

  • Transparency
  • Auditability
  • Custom health-specific decision making

It’s a foundation for user-defined weighting, personalized recommendations, and explainable AI in healthcare plan selection.


⚠️ Important note

This is an experimental demo, not medical or financial advice.
Always verify coverage details with the insurer.