Testimony summarization -- Just look at hearing_overview_pipeline.py

llm/hearing_overview_pipeline.py

@@ -0,0 +1,142 @@
+import json
+import numpy as np
+import os
+import pandas as pd
+import tiktoken
+import streamlit as st
+import urllib.request
+import chromadb
+import re
+import requests
+
+from chromadb.config import Settings
+from dataclasses import dataclass, field
+from langchain_text_splitters import TokenTextSplitter
+
+from operator import itemgetter
+from pathlib import Path
+from requests.exceptions import RequestException
+from sklearn.feature_extraction.text import TfidfVectorizer
+from sklearn.metrics.pairwise import cosine_similarity
+from typing import Tuple, List
+
+from prompts import *
+from tag_categories import *
+import firebase_admin
+from firebase_admin import credentials, firestore
+import openai
+import google.auth
+PROMPT_INSTRUCTIONS = "Follow these instructions when creating the prompt: \n" \
+"-Try to provide a balanced summary giving space to multiple sides of an issue that were shared in the hearing \n" \
+"-Try to avoid repeating uncritically the facts that were shared in testimony, as they may not be accurate \n" \
+"-Do not repeat any offensive, slanderous, or personally derogatory statements \n" \
+"-Note that the transcripts may contain transcription errors, such as mis-identified homophones, and that names referenced in the hearing may not have been transcribed accurately."
+
+# Set the environment variable for the Google Cloud project
+os.environ["GOOGLE_CLOUD_PROJECT"] = "digital-testimony-dev"
+
+@dataclass()
+class HearingDetails():
+    '''
+    A class to store all the details pertaining to a testimony. 
+    '''
+    hearing_id: str = ''
+    hearing_text: str = ''
+    summary: str = ''
+
+
+def recieve_hearing(hearing: HearingDetails):
+    hearing_text = hearing.hearing_text 
+    hearing_id = hearing.hearing_id
+    print(f'Receiving hearing with ID: {hearing_id}')
+    link = f'https://malegislature.gov/api/Hearings/{hearing_id}'
+    #Convert the hearing to a JSON object, allows us to access the bill ids
+    #that are mentioned during the hearing
+    r = requests.get(link, verify=False)
+    r = r.json()
+    bill_numbers = [
+    doc["BillNumber"]
+    for agenda in r.get("HearingAgendas", [])
+    for doc in agenda.get("DocumentsInAgenda", [])
+    ]
+    try:
+        db = connect_to_firestore()
+    except Exception as e:
+        print(f'Error connecting to Firestore: {e}')
+        st.error("Could not connect to Firestore. Please check your configuration.")
+        return hearing_text, {}
+
+    bill_summaries = {}
+    for number in bill_numbers:
+        # Get the bill document from Firestore
+        bill_ref = db.collection("generalCourts").document("194").collection("bills").document(number)
+        # Fetch the document
+        bill_doc = bill_ref.get()
+        if bill_doc.exists:
+            bill_data = bill_doc.to_dict()
+            # Append the summary to the list
+            bill_summaries.update({number: bill_data.get("summary", "")})
+        else:
+            st.warning(f"Bill {number} not found in Firestore.")
+    return hearing_text, bill_summaries
+
+def connect_to_firestore():
+    firebase_admin.initialize_app()
+    db = firestore.client()
+    return db
+
+def make_openai_request(prompt: str) -> str:
+    """
+    Make a request to OpenAI's API to get a response for the given prompt.
+    """
+    url = "https://api.openai.com/v1/chat/completions"
+
+    openai.api_key = os.environ.get("OPENAI_API_KEY")
+    headers = {
+        "Authorization": f"Bearer " + openai.api_key,
+        "Content-Type": "application/json"
+    }
+    data = {
+        "model": "gpt-4o-mini",
+        "messages": [
+            {"role": "user", "content": prompt}
+        ],
+    }
+    response = requests.post(url, headers=headers, json=data)
+    message = response.json()
+    return message['choices'][0]['message']['content']
+
+if __name__ == "__main__":
+    # Example usage
+    data = open('hearing-4548.json',)
+
+    testimony = json.load(data)
+
+    hearing_text = testimony['text']
+    hearing = HearingDetails(
+        hearing_id=4548,
+        hearing_text=hearing_text,
+        summary="This is a sample summary."
+    )
+    text, summaries = recieve_hearing(hearing)
+    PROMPT_BILL_SUMMARIES = f'''
+Provide a summary of this hearing to a regular person with no special knowledge or expertise of this area. 
+This is a hearing discussing several pending bills. 
+Provide a short summary of the sentiments that were expressed about the bills that were discussed during the hearing.
+Focus on which bills were discussed the most, and what the most common points were.
+Pull, if applicable, a compelling quote that is representative of a commonly made argument.
+If there was consensus on any specific point agreed upon by stakeholders who otherwise disagreed, please note that.
+You do not need to provide a summary of the bills themselves.
+Try and keep the overview of the hearing to 300 words or less.
+Follow these instructions when creating the prompt: {PROMPT_INSTRUCTIONS}
+The text of the hearing is as follows:
+```
+{text}
+```
+The summaries of each bill mentioned during the hearing are as follows:
+```
+{summaries}
+```
+'''
+    response = make_openai_request(PROMPT_BILL_SUMMARIES)
+    print(response)