Merge pull request #16 from shy982/functional-updates

Updates to processing logic & Documentation

README.md

@@ -29,9 +29,8 @@ To run the application (Production build deployment for demo purpose only, not a
 3. Wait a while for frontend and backend containers to spawn.
 4. Go to http://localhost:3000/ to start chatting
 
-# Notes:
+# Repository Handling Notes:
 
-1. Raise PR for Code updates (Anything to src directory)
-2. Push to main directly for MOM, References, Presentations, etc
-3. Read env.example to create .env for API tokens
-4. Mark TODO's as issues
+1. Raise PR on separate branch for code updates & request code owner review.
+2. Read env.example to create .env for API tokens.
+3. Mark TODO's as issues.

docs/README.md

@@ -1 +1,65 @@
 # Documentation of APIs, Services, and Modules
+
+# API Documentation
+
+## Overview
+This document outlines the available endpoints in the MediMate Q&A application. The application provides endpoints for processing user messages using different methods, including simple text processing, OpenAI's GPT-3.5 model, and a Retrieval-Augmented Generation (RAG) approach.
+
+## Endpoints
+
+### 1. Simple Chat Test
+- **Endpoint**: `/v1/chat/simple`
+- **Method**: `POST`
+- **Description**: A simple test endpoint that capitalizes the last user message. Used primarily for debugging purposes.
+- **Request Body**:
+  - `messages`: Array of message objects. Each message object should contain a `content` key with the message text.
+- **Response**: A string with the last message in uppercase.
+- **Example Request**:
+  ```json
+  {
+    "messages": [
+      {"content": "hello world"}
+    ]
+  }
+  ```
+
+### 2. Chat with GPT-3.5 (No RAG)
+- **Endpoint**: `/v1/chat/openai`
+- **Method**: `POST`
+- **Description**: Processes messages using OpenAI's GPT-3.5 model without RAG, considering the entire conversation history.
+- **Query Parameters**:
+  - `model` (optional): Specifies the GPT model to use. Default is "gpt-3.5-turbo-instruct".
+- **Request Body**:
+  - `messages`: Array of message objects as described above.
+  - `medicalHistory` (optional): String containing the medical history to be considered in the response.
+- **Response**: A string containing the AI-generated response.
+- **Example Request**:
+  ```json
+  {
+    "messages": [
+      {"content": "What are the symptoms of a cold?"}
+    ],
+    "medicalHistory": "Patient has a history of allergies."
+  }
+  ```
+
+### 3. Chat with RAG
+- **Endpoint**: `/v1/chat/openai/rag`
+- **Method**: `POST`
+- **Description**: Processes messages using a RAG approach, combining GPT-3.5 and external knowledge retrieval.
+- **Query Parameters**:
+  - `model` (optional): GPT model to use. Default is "gpt-3.5-turbo-instruct".
+  - `dataset` (optional): The dataset for knowledge retrieval. Default is "nfcorpus".
+- **Request Body**:
+  - `messages`: Array of message objects as described above.
+  - `medicalHistory` (optional): String containing relevant medical history.
+- **Response**: A string containing the AI-generated response with external knowledge context.
+- **Example Request**:
+  ```json
+  {
+    "messages": [
+      {"content": "Tell me about diabetes management."}
+    ],
+    "medicalHistory": "Patient has type 2 diabetes."
+  }
+  ```

src/main/backend/qna_service/openai_client.py

@@ -9,8 +9,13 @@
 from langchain.schema.output_parser import StrOutputParser
 from langchain.schema.runnable import RunnablePassthrough
 from langchain.vectorstores import FAISS
+import re
 
 
+def clean_response(response):
+    cleaned_response = re.sub(r'^\W+', '', response)
+    return cleaned_response
+
 def request_gpt_no_rag(messages, medical_history, model):
     load_dotenv()
     client = openai.OpenAI()
@@ -25,7 +30,7 @@ def request_gpt_no_rag(messages, medical_history, model):
         prompt="Answer the last question of this conversation briefly: " + conversation_history,
         max_tokens=200
     )
-    return answer.choices[0].text.strip()
+    return clean_response(answer.choices[0].text.strip()) if answer!= '' else "OpenAI API temporarily down :( Please try again in a while."
 
 
 def run_rag_pipeline(messages, medical_history, model="gpt-3.5-turbo-instruct", dataset="nfcorpus"):
@@ -68,4 +73,4 @@ def format_docs(_docs):
     # Run the RAG pipeline
     response = chain.invoke(conversation_history)
 
-    return response.strip()
+    return clean_response(response.strip()) if response!= '' else "OpenAI API temporarily down :( Please try again in a while."