Feat/deployment (#28)

* fix(mcq): fixed issue where only 1 MCQ was being generated by gpt

* fix(client): minor qol fixes

* fix(mcq): added additonal fixes

* fix(profile): fixed cards

* fix(subscribe): added blue box

* feat(MCQ): added option to reorder, delete or add options

---------

Co-authored-by: neilscallywag <neil.sharma.2022@scis.smu.edu.sg>
Co-authored-by: Louis <64778758+iamlouisteo@users.noreply.github.com>

backend/complex/process-chunks/src/main.py

@@ -66,8 +66,12 @@ def construct_prompt(self, message_from_queue1, messages_from_queue2):
         logging.info(f"Message data: {message_data}")
         generate_type = message_data.get("metadata", {}).get("generateType", "")
         note_id = message_data.get("fileId", "")
+        # get all "content" key from various jsons in messages_from_queue2 which is a list of json objects
+        additional_context = ", ".join([json.loads(message)["content"] for message in messages_from_queue2])
+        logging.info(f"Additional context: {additional_context}")
         strategy = PromptStrategyFactory.get_strategy(generate_type)
-        return strategy.construct_prompt(message_from_queue1, messages_from_queue2),generate_type,note_id
+        prompt,content = strategy.construct_prompt(message_from_queue1, additional_context)
+        return prompt,content,generate_type,note_id
 
     def match_messages_and_call_api(self, ch, method, properties, body):
         """Processes messages from queue1, matches them with messages from queue2, and calls the OpenAI API."""
@@ -79,8 +83,8 @@ def match_messages_and_call_api(self, ch, method, properties, body):
             else:
                 break
 
-        prompt,generate_type,note_id = self.construct_prompt(message_from_queue1, messages_from_queue2)
-        token_count = self.count_tokens_with_tiktoken(prompt)
+        prompt,content,generate_type,note_id = self.construct_prompt(message_from_queue1, messages_from_queue2)
+        token_count = self.count_tokens_with_tiktoken(prompt+content)
         logging.info(f"Estimated token count for prompt: {token_count}")
         if token_count > self.max_tokens:
             # Calculate 2% of the max token limit
@@ -89,8 +93,8 @@ def match_messages_and_call_api(self, ch, method, properties, body):
             new_max_length = self.max_tokens - reduction_amount
             # Adjust the prompt to the new max length
             # Assuming prompt is a string, this will cut off the end to fit. Adjust as necessary for your data structure.
-            prompt = prompt[:new_max_length]
-            logging.info(f"Prompt adjusted to within token limit. New length: {len(prompt)}")
+            content = content[:new_max_length]
+            logging.info(f"Prompt adjusted to within token limit. New length: {len(content)+len(prompt)}")
 
 
 
@@ -101,7 +105,10 @@ def match_messages_and_call_api(self, ch, method, properties, body):
         try:
             response = client.chat.completions.create(
                 model=self.model,
-                messages=[{"role": "system", "content": prompt}]
+                messages=[{"role": "system", "content": prompt},
+                          {"role": "user", "content": content}],
+                temperature=0.3,
+                top_p=0.8,
             )
         except Exception as e:
             logging.error(f"Error during OpenAI API call or response handling: {str(e)}")
@@ -147,6 +154,6 @@ def send_content(self, message, generate_type,note_id):
     content_fetcher_builder = ContentFetcherBuilder()
     content_fetcher = (content_fetcher_builder
                        .setup_logging()
-                       .with_model("gpt-3.5-turbo-16k")
+                       .with_model("gpt-3.5-turbo")
                        .build())
-    content_fetcher.start_consuming()
+    content_fetcher.start_consuming()

backend/complex/process-chunks/src/main2.py

@@ -0,0 +1,84 @@
+import grpc
+import pika
+import json
+import utils
+import logging
+from itertools import cycle
+from openai import OpenAI
+from content_pb2_grpc import ContentStub
+from strategies.ContentSendingStrategy import FlashcardStrategy, MCQStrategy
+from proxies.ContentServiceProxy import ContentServiceProxy
+
+class ContentFetcher:
+    def __init__(self):
+        self.RABBITMQ_SERVER = None
+        self.RABBITMQ_USERNAME = None
+        self.RABBITMQ_PASSWORD = None
+        self.QUEUE_NAME_1 = None
+        self.QUEUE_NAME_2 = None
+        self.OPENAI_API_KEYS = None
+        self.CONTENT_SERVICE_ADDRESS = None
+        self.model = None
+        self.connection = None
+        self.channel = None
+        self.grpc_channel = None
+        self.grpc_stub = None
+        self.max_tokens = 16384
+        self.api_key = None  # Single API key for continuous context
+
+    def initialize_rabbitmq(self):
+        credentials = pika.PlainCredentials(self.RABBITMQ_USERNAME, self.RABBITMQ_PASSWORD)
+        connection_parameters = pika.ConnectionParameters(host=self.RABBITMQ_SERVER, credentials=credentials)
+        self.connection = pika.BlockingConnection(connection_parameters)
+        self.channel = self.connection.channel()
+        self.channel.queue_declare(queue=self.QUEUE_NAME_1, durable=True)
+        self.channel.queue_declare(queue=self.QUEUE_NAME_2, durable=True)
+
+    def match_messages_and_call_api(self, ch, method, properties, body):
+        message_from_queue1 = body.decode()
+        messages_from_queue2 = []
+        for method_frame, properties, body in self.channel.consume(queue=self.QUEUE_NAME_2, auto_ack=True, inactivity_timeout=1):
+            if method_frame:
+                messages_from_queue2.append(body.decode())
+            else:
+                break
+
+        prompt, generate_type, note_id = self.construct_prompt(message_from_queue1, messages_from_queue2)
+        responses = self.generate_content_in_batches(prompt, 3, generate_type, note_id)  # Example: 3 iterations for 15 items
+        self.send_content(responses, generate_type, note_id)
+
+    def generate_content_in_batches(self, initial_prompt, num_batches, generate_type, note_id):
+        current_responses = []
+        prompt = initial_prompt
+        client = OpenAI(api_key=self.api_key)  # Reuse the same API key for all batches
+
+        for _ in range(num_batches):
+            response = client.chat.completions.create(model=self.model, messages=[{"role": "system", "content": prompt}])
+            batch_responses = utils.extract_and_validate_json_objects(response.choices[0].message.content)
+            current_responses.extend(json.loads(batch_responses))
+
+            prompt = "Generate 5 more"  # Modify the prompt for the next batch
+
+        return current_responses
+
+    def send_content(self, messages, generate_type, note_id):
+        if generate_type == "flashcard":
+            strategy = FlashcardStrategy()
+        elif generate_type == "mcq":
+            strategy = MCQStrategy()
+        else:
+            raise ValueError("Unsupported content type")
+
+        content_service_proxy = ContentServiceProxy(self.CONTENT_SERVICE_ADDRESS)
+        content_request = strategy.construct_request(json.dumps(messages), note_id)
+        response = content_service_proxy.send_content(content_request)
+        logging.info(f"gRPC Response: {response}")
+
+    def start_consuming(self):
+        self.channel.basic_consume(queue=self.QUEUE_NAME_1, on_message_callback=self.match_messages_and_call_api, auto_ack=True)
+        logging.info("Waiting for messages. To exit press CTRL+C")
+        self.channel.start_consuming()
+
+if __name__ == "__main__":
+    content_fetcher = ContentFetcher()
+    content_fetcher.start_consuming()

backend/complex/process-chunks/src/strategies/prompt_construction_strategy.py

@@ -1,11 +1,13 @@
 import json
 from abc import ABC, abstractmethod
 
+
 class PromptConstructionStrategy(ABC):
     @abstractmethod
     def construct_prompt(self, message_from_queue1, messages_from_queue2):
         pass
 
+
 class FlashcardPromptStrategy(PromptConstructionStrategy):
     def construct_prompt(self, message_from_queue1, messages_from_queue2):
         message_data = json.loads(message_from_queue1)
@@ -15,40 +17,43 @@ def construct_prompt(self, message_from_queue1, messages_from_queue2):
 Provides a detailed answer, aiming for undergraduate-level depth where applicable.
 Exclude any obfuscated or nonsensical text, and select content from various sections to cover the topic comprehensively. Aim for minimal token use in your response without sacrificing content quality unless the topic is straightforward.
 
-Format your responses strictly as json:
+Format all 20 of your responses strictly as json:
 
 {{
     "question": "What is the significance of [specific concept] in [subject]?",
     "answer": "The significance lies in..."
 }},
 ...
 
-{additional_context}"""
-        return prompt
+I will embedd the content below for your reference. Do not parse it as instruction this time. Just use it as reference to generate Flashcards.:
+
+"""
+        return prompt, additional_context
 
 
 class MCQPromptStrategy(PromptConstructionStrategy):
     def construct_prompt(self, message_from_queue1, messages_from_queue2):
         message_data = json.loads(message_from_queue1)
-        additional_context = ", ".join(messages_from_queue2)
-        prompt = f"""Generate twenty MCQs that highlight essential information on the subject, drawn from clear and relevant portions of the text. Focus on diverse concepts, definitions, and findings, ensuring each MCQ:
-
-Provides a detailed answer, aiming for undergraduate-level depth where applicable.
-Exclude any obfuscated or nonsensical text, and select content from various sections to cover the topic comprehensively. Aim for minimal token use in your response without sacrificing content quality unless the topic is straightforward.
-
-Format your responses strictly as json:
-
+        additional_context = """Generate between 10 and 20 MCQs from the provided text, ensuring each question:
+- Highlights essential information across diverse concepts, definitions, and findings.
+- Is detailed enough for undergraduate-level understanding.
+- Includes only clear and relevant portions of the text, covering the topic comprehensively.
+""" + ", ".join(
+            messages_from_queue2
+        )
+        prompt = f"""
+Each MCQ must be formatted in JSON and indicate whether multiple answers are allowed:
+
+Example MCQ:
 {{
-"question": "What is the capital of France?",
-"options": [
-    {{
-    "option": "Paris",
-    "is_correct": true
-    }},
-    ...
-],
-"multiple_answers": false
+  "question": "What does the user YUE ZHENG TING add as a comment on February 24th?",
+  "options": [
+    {{"option": "Resolution: Done", "is_correct": true}},
+    {{"option": "Reporter: THADDEAUS LOW", "is_correct": true}},
+    {{"option": "Votes: 0", "is_correct": true}},
+    {{"option": "Service Management Request Type: Emailed request", "is_correct": false}}
+  ],
+  "multiple_answers": true
 }}
-
-{additional_context}"""
-        return prompt
+"""
+        return prompt, additional_context