Skip to content
Start a project
Start a project

[email protected]LinkedIn

Client projectClient confidential

A chatbot that answers healthcare questions from a medical knowledge base.

Services
AI Engineering
Industry
Healthcare
Client
Confidential
Built with
Python · LangChain · FAISS · Hugging Face · Streamlit · PostgreSQL
The assistant's landing page: trusted medical answers from a knowledge base, with the sources behind each one.

01

General chatbots can give confident answers with no source behind them. In healthcare, every answer needs to come from reliable material.

02

  1. 01Retrieval-augmented generation (RAG): documents are embedded and searched, and the most relevant passages go to the model with each question.
  2. 02LangChain to orchestrate retrieval and the model.
  3. 03A retrieval workflow designed to reduce unsupported answers.
  4. 04A Streamlit interface for asking questions in conversation.
Built with
  • Python 3.12
  • LangChain
  • FAISS (CPU)
  • Hugging Face Inference API
  • Sentence Transformers
  • Embedding models
  • Reranking (reranker model)
  • LLMs
  • Vector database
  • PyPDF
  • OCR
  • PostgreSQL
  • Gmail SMTP
  • Streamlit
  • Flask
  • React
  • Vue.js
  • Bootstrap
  • HTML
  • CSS
  • JavaScript

The product

A question about type 2 diabetes answered from the knowledge base, with symptoms, diagnosis criteria and the sources used.
The knowledge base: medical guidelines and papers with their processing status and an AI summary of each.
Chat history: past medical questions with their topics and dates.

Representative interface; the client's product is confidential.

Next project

FetchLead AI

Read the case study