Inspiration
Leaving hospital can mean taking home a lot of information at once. Medication instructions, follow-up appointments and care guidance may be spread across several pages. We wanted to make that information easier for patients and carers to find and understand.
That idea became Discharge Bridge: a bridge between written discharge instructions and the questions people have when they get home.
What it does
Discharge Bridge lets users upload hospital discharge PDFs and ask questions in plain language, such as:
- “When is my follow-up appointment?”
- “What medication instructions are listed?”
- “What does the document say about my dressing?”
The app retrieves relevant passages and generates an explanation with source references. Users can expand the original excerpts and check the document name and page number.
When the documents do not support an answer, the app is instructed to acknowledge the missing information. It is designed to explain documented instructions, rather than diagnose conditions or recommend treatment changes.
How we built it
We built the interface with Streamlit and used pypdf to extract text from each PDF page.
The app divides that text into overlapping chunks, retaining the document name and page number. A Hugging Face SentenceTransformer converts the chunks into embeddings, and FAISS searches for passages relevant to each question.
The retrieved passages and question are sent to an OpenAI language model, which is instructed to answer from those passages and include source citations. The app checks the response format and citation IDs before displaying it.
Built With
- openai
- python
