feat(integration): multilingual-agent

integrations/multilingual-agent/project.json

@@ -0,0 +1,6 @@
+{
+    "title": "Multilingual-agent",
+    "description": "Agent with multilingual capabilities to summarize a youtube video, translate the summary to a different language, and generate a text response in the target language.",
+    "categories":    ["Text Summarization", "Hugging Face"], 
+    "deltav": false
+}

integrations/multilingual-agent/pyproject.toml

@@ -0,0 +1,18 @@
+[tool.poetry]
+name = "multi-lingual-video-agent"
+version = "0.1.0"
+description = "Agent that can summarize a video in different languages from english"
+authors = ["Lenin Kennedy <[email protected]>", "Samuela Abigail Mathew <[email protected]>"]
+license = "MIT License"
+readme = "README.md"
+
+[tool.poetry.dependencies]
+python = ">=3.10,<3.12"
+uagents = "0.11.0"
+youtube-transcript-api = "^0.6.2"
+langchain = "^0.1.12"
+
+
+[build-system]
+requires = ["poetry-core"]
+build-backend = "poetry.core.masonry.api"

integrations/multilingual-agent/src/agents/multilingual_agent.py

@@ -0,0 +1,97 @@
+# Necessary imports: uagents for agent creation and message handling,
+# os and requests for managing API calls
+from uagents import Agent, Context, Protocol, Model
+from pydantic import Field
+from ai_engine import UAgentResponse, UAgentResponseType
+from messages.basic import UAResponse, UARequest, Error
+from uagents.setup import fund_agent_if_low
+import os
+import requests
+from utils.functions import get_video_script, summarize_transcript
+
+# The access token and URL for the SAMSUM BART model, served by Hugging Face
+HUGGING_FACE_ACCESS_TOKEN = os.getenv(
+    "HUGGING_FACE_ACCESS_TOKEN", "HUGGING FACE secret phrase :)")
+SAMSUM_BART_URL = "https://api-inference.huggingface.co/models/Samuela39/my-samsum-model"
+
+# Setting the headers for the API call
+HEADERS = {
+    "Authorization": f"Bearer {HUGGING_FACE_ACCESS_TOKEN}"
+}
+
+SEED=HUGGING_FACE_ACCESS_TOKEN
+
+
+# Copy the address shown below
+print(f"Your agent's address is: {Agent(seed=SEED).address}")
+
+
+# Then go to https://agentverse.ai, register your agent in the Mailroom
+# and copy the agent's mailbox key
+AGENT_MAILBOX_KEY = os.getenv("AGENT_MAILBOX_KEY", None)
+assert (
+    AGENT_MAILBOX_KEY is not None
+), "AGENT_MAILBOX_KEY environment variable is not set! Please set it to your agent's mailbox key!"
+
+# Now your agent is ready to join the agentverse!
+agent = Agent(
+    name="multilingual-agent",
+    seed=SEED,
+    mailbox=f"{AGENT_MAILBOX_KEY}@https://agentverse.ai",
+)
+
+# # Creating the agent and funding it if necessary
+# agent = Agent(
+#     name="multilingual-agent",
+#     seed=SEED,
+#     port=8001,
+#     endpoint=["http://127.0.0.1:8001/submit"],
+# )
+# fund_agent_if_low(agent.wallet.address())
+
+class UAResponse(Model):
+    response: list
+
+class SummarizationRequest(Model):
+    url: str = Field(description="URL of the video you want to summarize")
+
+# Protocol declaration for UARequests
+multilingual_agent = Protocol("summary-request","0.1.0")
+
+# Declaration of a message event handler for handling UARequests and send respective response.
+@multilingual_agent.on_message(model=SummarizationRequest, replies={UAResponse, Error})
+async def handle_request(ctx: Context, sender: str, msg: SummarizationRequest):
+    # Logging the request information
+    ctx.logger.info(
+        f"Got request from  {sender} for summarization : {msg.url}")
+
+    try:
+        transcript = get_video_script(msg.video_link)
+    except Exception as e:
+        await ctx.send(
+            sender,
+            UAgentResponse(
+                message=f"No transcript found for video! Error: {e}",
+                type=UAgentResponseType.ERROR,
+            ),
+        )
+        return
+
+    summary = await summarize_transcript(transcript, msg.model)
+
+    print("Sending short article")
+    await ctx.send(
+        sender,
+        UAgentResponse(
+            message=f"Summary:\n{summary}",
+            type=UAgentResponseType.FINAL,
+        ),
+    )
+    return
+
+
+# Include protocol to the agent
+agent.include(multilingual_agent, publish_manifest=True)
+
+if __name__ == "__main__":
+    agent.run()

integrations/multilingual-agent/src/messages/basic.py

@@ -0,0 +1,13 @@
+from uagents import Model
+
+
+class UARequest(Model):
+    text: str
+
+
+class Error(Model):
+    error: str
+
+
+class UAResponse(Model):
+    response: list

integrations/multilingual-agent/src/utils/functions.py

@@ -0,0 +1,91 @@
+from urllib.parse import urlparse, parse_qs
+from youtube_transcript_api import YouTubeTranscriptApi
+from langchain_text_splitters import RecursiveCharacterTextSplitter
+import os
+import requests
+
+# The access token and URL for the SAMSUM BART model, served by Hugging Face
+HUGGING_FACE_ACCESS_TOKEN = os.getenv(
+    "HUGGING_FACE_ACCESS_TOKEN", "HUGGING FACE secret phrase :)")
+SAMSUM_BART_URL = "https://api-inference.huggingface.co/models/Samuela39/my-samsum-model"
+
+# Setting the headers for the API call
+HEADERS = {
+    "Authorization": f"Bearer {HUGGING_FACE_ACCESS_TOKEN}"
+}
+
+
+def get_video_script(url: str) -> list:
+    """
+    Get the script of a YouTube video by its URL and return it as a list of strings
+    """
+
+    if not url:
+        raise ValueError("YouTube video url is required")
+
+    video_url = urlparse("https://www.youtube.com/watch?v=1c9iyoVIwDs")
+    video_query = parse_qs(video_url.query)
+
+    video_id = None
+
+    if "v" in video_query:
+        video_id = video_query["v"][0]
+    else:
+        raise ValueError("Invalid YouTube video url")
+
+    if not video_id:
+        raise ValueError("Invalid YouTube video url")
+
+    video_script = YouTubeTranscriptApi.get_transcript(video_id, languages=["en"])
+
+    video_text = ""
+
+    for segment in video_script:
+        video_text += segment['text'] + " "
+
+    return [segment['text'] for segment in video_script if segment["text"]]
+
+def chunk_text(text: list, chunk_size: int = 1000, chunk_overlap: int = 100) -> list:
+    """
+    Split a list of strings into chunks of strings
+    """
+    text_splitter = RecursiveCharacterTextSplitter(
+        chunk_size=chunk_size,
+        chunk_overlap=chunk_overlap,
+        length_function=len,
+    )
+
+    texts = text_splitter.create_documents(text)
+
+    return texts
+
+def get_summarization(text: str) -> str:
+    """
+    Summarize a string
+    """
+    data = {
+        "inputs": text
+    }
+
+    response = requests.post(SAMSUM_BART_URL, headers=HEADERS, json=data)
+    model_res = response.json()[0]
+    summary = model_res['summary_text']
+    return summary
+
+def summarize_transcript(text: list, chunk_size: int = 1000, chunk_overlap: int = 100) -> list:
+    """
+    Summarize a list of strings
+    """
+    chunks = chunk_text(text, chunk_size, chunk_overlap)
+
+    summarized_chunks = []
+
+    for chunk in chunks:
+        summarized_text = get_summarization(chunk)
+        summarized_chunks.append(summarized_text)
+
+    summarized_text =''
+    for summary in summarized_chunks:
+        summarized_text += get_summarization(summary)
+
+    return summarized_text