Spaces:
Sleeping
Sleeping
File size: 8,342 Bytes
1fac056 10ff263 1fac056 10ff263 1fac056 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 | from agents import Agent, Runner
from composio_openai_agents import ComposioToolSet, App, Action
import streamlit as st
from dotenv import load_dotenv
import asyncio
from datetime import datetime, timedelta
import pandas as pd
import plotly.express as px
from wordcloud import WordCloud
import matplotlib.pyplot as plt
from io import BytesIO
import base64
from pydantic import BaseModel
from typing import List, Optional
import os
load_dotenv()
max_date = datetime.now().strftime("%Y-%m-%d")
min_date = (datetime.now() - timedelta(days=30)).strftime("%Y-%m-%d")
class SentimentDatum(BaseModel):
sentiment: str
count: int
class IncidentDatum(BaseModel):
source: str
incidents: int
class CompetitiveDatum(BaseModel):
company: str
mentions: int
class Citation(BaseModel):
source: str
text: str
url: str
sentiment: str
class DashboardOutput(BaseModel):
executive_summary: str
sentiment_data: List[SentimentDatum]
incident_data: List[IncidentDatum]
competitive_data: List[CompetitiveDatum]
wordcloud_text: str
citations: List[Citation]
openai_key = os.environ.get("OPENAI_API_KEY")
composio_key = os.environ.get("COMPOSIO_API_KEY")
async def run_analysis(company_name):
toolset = ComposioToolSet(api_key=composio_key)
tools = toolset.get_tools(actions=[Action.TWITTER_RECENT_SEARCH, Action.REDDIT_SEARCH_ACROSS_SUBREDDITS, Action.HACKERNEWS_SEARCH_POSTS, Action.HACKERNEWS_GET_ITEM_WITH_ID])
agent = Agent(
name="Social Media Analyst",
instructions=f"""
Your ONLY job is to deliver a crisp, actionable, clustered executive summary of new, high-utility, real-time technical incidents, security chatter, actionable customer/developer feedback for the company, and competitive intelligence.
- Do NOT cover general market or sector sentiment, financial/market positioning, or broad sentiment.
- Do NOT explain what you are not doing, do NOT ask for clarification, do NOT disclaim, and do NOT ask for permission or confirmation.
- Just deliver the report as specified, following the structure below.
- You must use Reddit, Twitter, and Hacker News to get the data. For each source, find and cluster at least 3 relevant, recent signals (if available) and include them in the summary and citations.
- The search query for twitter has to be something like this (Adjust the parameters according to the context): "{company_name} to:{max_date} from:{min_date}", don't pass any other fields. Just the query and max results.
- For Twitter: If no technical incidents, bugs, outages, or actionable feedback are found, instead summarize the general sentiment about {company_name} on Twitter, with clickable links to the original tweets.
- For hackernews, use the search query: "{company_name}", then use the get item with id tool to get the item.
- Don't give updates that the company already knows about, like product launches, new features, etc.
- For competitive intelligence, look for posts, tweets, or comments where users compare {company_name} to other competitors. Summarize the key points, advantages, disadvantages, and user sentiment in these comparisons. Include citations for each comparison.
- Provide more research-detailed information in both the executive summary and competitive intelligence sections. Go deeper into technical, user, and product details, and cite specific sources and findings.
Report structure:
1. # Executive Summary: A crisp, actionable, clustered summary of all new, high-utility signals. No fluff, only what matters. Include detailed research and findings.
2. # Competitive Intelligence: What users are saying when comparing {company_name} to competitors. Summarize key themes, pros/cons, and user sentiment. Include citations and research-level detail.
3. # Exact Citations: For each cluster or item, provide clickable links to the original sources (tweets, posts, etc.).
Everything should be cited and referenced. Focus only on the tech side and actionable intelligence.
You will be given the name of the company.
If tool call fails, try again with different permutations of the query.
""",
tools=tools,
model='gpt-4.1',
)
res = await Runner.run(starting_agent=agent, input=f"What is the summary of {company_name}?")
return res.final_output
async def dashboard_result(result):
agent = Agent(
name="Dashboard Generator",
instructions=f"""
Your job is to generate a dashboard from the given result.
Competitors include direct competitors only.
""",
output_type=DashboardOutput,
)
res = await Runner.run(starting_agent=agent, input=result)
return res.final_output
name_from_url = st.query_params.get("name", "")
company_from_url = st.query_params.get("company", "")
if name_from_url:
st.title(f"{name_from_url}'s Market Pulse Agent ")
else:
st.title("Market Pulse Agent")
company_name = st.text_input("Company name:", value=company_from_url)
if st.button("Analyze"):
if company_name:
with st.spinner("Analyzing social media sentiment..."):
result = asyncio.run(run_analysis(company_name))
dashboard_result_response = asyncio.run(dashboard_result(str(result)))
st.subheader("Summary and Sentiment Analysis")
st.write(result)
st.markdown("---")
st.header("📊 Dashboard")
# Metrics Row
metric1, metric2, metric3 = st.columns(3)
with metric1:
st.metric("Total Mentions", sum([s.count for s in dashboard_result_response.sentiment_data]))
with metric2:
st.metric("Incidents", sum([i.incidents for i in dashboard_result_response.incident_data]))
with metric3:
st.metric("Competitors Compared", len(dashboard_result_response.competitive_data)-1)
# Main Grid
grid1, grid2 = st.columns([2, 1])
with grid1:
st.subheader(":green[Sentiment Breakdown]")
df_sentiment = pd.DataFrame([s.model_dump() for s in dashboard_result_response.sentiment_data])
fig = px.pie(df_sentiment, names='sentiment', values='count', color='sentiment',
color_discrete_map={"Positive": "#22c55e", "Negative": "#ef4444", "Neutral": "#a3a3a3"})
fig.update_traces(textinfo='percent+label', pull=[0.05, 0.05, 0.05])
st.plotly_chart(fig, use_container_width=True)
# Competitive Intelligence full width
st.subheader(":blue[Competitive Intelligence]")
df_comp = pd.DataFrame([c.model_dump() for c in dashboard_result_response.competitive_data])
fig3 = px.bar(df_comp, x='company', y='mentions', color='company', text='mentions',
color_discrete_sequence=px.colors.qualitative.Pastel)
fig3.update_layout(yaxis_title=None, xaxis_title=None, plot_bgcolor='#f8fafc')
st.plotly_chart(fig3, use_container_width=True)
with grid2:
if sum([i.incidents for i in dashboard_result_response.incident_data]) > 0:
st.subheader(":orange[Incident Frequency]")
df_incident = pd.DataFrame([i.model_dump() for i in dashboard_result_response.incident_data])
fig2 = px.bar(df_incident, x='source', y='incidents', color='source', text='incidents',
color_discrete_sequence=px.colors.qualitative.Set2)
fig2.update_layout(yaxis_title=None, xaxis_title=None, plot_bgcolor='#f8fafc')
st.plotly_chart(fig2, use_container_width=True)
st.markdown("<hr style='margin:2rem 0;' />", unsafe_allow_html=True)
st.subheader(":gray[Top Citations]")
df_cite = pd.DataFrame([c.model_dump() for c in dashboard_result_response.citations])
def make_clickable(url):
return f'<a href="{url}" target="_blank" style="color:#2563eb; text-decoration:underline;">link</a>'
df_cite['url'] = df_cite['url'].apply(make_clickable)
st.write(df_cite.to_html(escape=False, index=False, classes='tw-table tw-table-auto tw-bg-white tw-shadow-md tw-rounded-lg'), unsafe_allow_html=True)
else:
st.warning("Please enter a company name.") |