import streamlit as st
import time
import re
import io
from datetime import datetime
from fpdf import FPDF
from dotenv import load_dotenv
from src.graph import app
from src.state import AgentState
# Load environment variables (API Keys) from .env
load_dotenv()
st.set_page_config(page_title="Startup Validator", page_icon="π", layout="wide")
# ββββββββββββββββββββββββββββββββββββββββββββββ
# INIT CACHE
# ββββββββββββββββββββββββββββββββββββββββββββββ
if "cache" not in st.session_state:
st.session_state.cache = {}
# ββββββββββββββββββββββββββββββββββββββββββββββ
# PDF REPORT GENERATOR
# ββββββββββββββββββββββββββββββββββββββββββββββ
def _clean(text: str) -> str:
"""Strip markdown and non-latin chars for PDF."""
text = re.sub(r'\*\*(.+?)\*\*', r'\1', text)
text = re.sub(r'\*(.+?)\*', r'\1', text)
return text.encode("latin-1", "replace").decode("latin-1")
def generate_pdf(idea, score, verdict, summary, all_results):
pdf = FPDF()
pdf.set_auto_page_break(auto=True, margin=15)
pdf.add_page()
# ββ Title ββ
pdf.set_font("Helvetica", "B", 26)
pdf.set_text_color(55, 48, 107)
pdf.cell(0, 16, "Startup Validator Report", ln=True, align="C")
pdf.set_font("Helvetica", "", 9)
pdf.set_text_color(140, 140, 140)
pdf.cell(0, 7, f"Generated on {datetime.now().strftime('%B %d, %Y at %I:%M %p')}", ln=True, align="C")
pdf.ln(6)
# ββ Thin line ββ
pdf.set_draw_color(200, 200, 220)
pdf.line(15, pdf.get_y(), 195, pdf.get_y())
pdf.ln(8)
# ββ Startup Idea ββ
pdf.set_text_color(40, 40, 40)
pdf.set_font("Helvetica", "B", 13)
pdf.cell(0, 9, "Startup Idea", ln=True)
pdf.set_font("Helvetica", "", 11)
pdf.multi_cell(0, 6.5, _clean(idea))
pdf.ln(6)
# ββ Score & Verdict ββ
pdf.set_font("Helvetica", "B", 13)
pdf.cell(0, 9, "Evaluation", ln=True)
# Colored verdict box
if verdict == "Go":
pdf.set_fill_color(5, 150, 105)
elif verdict == "Caution":
pdf.set_fill_color(217, 119, 6)
else:
pdf.set_fill_color(220, 38, 38)
pdf.set_font("Helvetica", "B", 14)
pdf.set_text_color(255, 255, 255)
pdf.cell(50, 10, f" Score: {score}/100", fill=True, new_x="RIGHT", new_y="TOP")
pdf.cell(5, 10, "")
pdf.cell(40, 10, f" {verdict}", fill=True, ln=True)
pdf.ln(6)
# ββ Executive Summary ββ
pdf.set_text_color(40, 40, 40)
pdf.set_font("Helvetica", "B", 13)
pdf.cell(0, 9, "Executive Summary", ln=True)
pdf.set_font("Helvetica", "", 10.5)
pdf.multi_cell(0, 6, _clean(summary))
pdf.ln(6)
pdf.set_draw_color(200, 200, 220)
pdf.line(15, pdf.get_y(), 195, pdf.get_y())
pdf.ln(6)
# ββ Agent Reports ββ
pdf.set_font("Helvetica", "B", 15)
pdf.set_text_color(55, 48, 107)
pdf.cell(0, 10, "Detailed Agent Reports", ln=True)
pdf.ln(3)
agent_labels = {
"competitor_analysis": "Competitor Analysis",
"market_sizing": "Market Sizing",
"community_sentiment": "Community Sentiment",
"os_risk": "Open-Source Risk",
}
for result in all_results:
agent_key = result.get("agent", "unknown")
label = agent_labels.get(agent_key, agent_key.replace("_", " ").title())
findings = result.get("findings", "No findings.")
sources = result.get("sources", [])
# Agent heading
pdf.set_font("Helvetica", "B", 12)
pdf.set_text_color(60, 60, 60)
pdf.cell(0, 8, label, ln=True)
# Findings
pdf.set_font("Helvetica", "", 10)
pdf.set_text_color(70, 70, 70)
pdf.multi_cell(0, 5.5, _clean(findings))
# Sources
if sources:
pdf.set_font("Helvetica", "I", 8.5)
pdf.set_text_color(90, 90, 180)
for src in sources:
link_text = _clean(f" [{src.get('domain', '')}] {src.get('title', '')[:60]}")
pdf.cell(0, 4.5, link_text, ln=True, link=src.get("url", ""))
pdf.set_text_color(70, 70, 70)
pdf.ln(5)
# ββ Footer ββ
pdf.ln(8)
pdf.set_font("Helvetica", "I", 8)
pdf.set_text_color(160, 160, 160)
pdf.cell(0, 5, "Generated by Startup Validator - AI Multi-Agent System", ln=True, align="C")
return bytes(pdf.output())
# ββββββββββββββββββββββββββββββββββββββββββββββ
# PREMIUM CSS
# ββββββββββββββββββββββββββββββββββββββββββββββ
st.markdown("""
""", unsafe_allow_html=True)
# ββββββββββββββββββββββββββββββββββββββββββββββ
# HERO HEADER
# ββββββββββββββββββββββββββββββββββββββββββββββ
st.markdown("""
π
Startup Validator
AI-Powered Multi-Agent Due Diligence β in Seconds
π Competitor Analysis
π Market Sizing
π¬ Community Sentiment
β οΈ Open-Source Risk
""", unsafe_allow_html=True)
# ββββββββββββββββββββββββββββββββββββββββββββββ
# INPUT
# ββββββββββββββββββββββββββββββββββββββββββββββ
idea = st.text_area(
"π‘ Describe your startup idea",
height=120,
placeholder="e.g., A marketplace for renting out private swimming pools by the hourβ¦",
)
if st.button("β‘ Validate My Idea", type="primary"):
if not idea.strip():
st.warning("Please enter an idea first.")
else:
st.markdown('', unsafe_allow_html=True)
idea_key = idea.strip().lower()
is_cached = idea_key in st.session_state.cache
# ββ Agent metadata ββ
agent_meta = {
"competitor_agent": {"icon": "π", "label": "Competitor Analysis"},
"market_sizing_agent": {"icon": "π", "label": "Market Sizing"},
"sentiment_agent": {"icon": "π¬", "label": "Community Sentiment"},
"os_risk_agent": {"icon": "β οΈ", "label": "Open-Source Risk"},
"scoring_node": {"icon": "π§ ", "label": "Scoring Engine"},
}
if is_cached:
# ββββββββββββββββββββββββββββββββββ
# CACHED RESULT β instant display
# ββββββββββββββββββββββββββββββββββ
cached = st.session_state.cache[idea_key]
final_state = cached["final_state"]
all_results = cached["all_results"]
elapsed = cached["elapsed"]
st.markdown(
''
'β‘ Cached result β instant load
',
unsafe_allow_html=True,
)
else:
# ββββββββββββββββββββββββββββββββββ
# FRESH RUN β stream with progress bar
# ββββββββββββββββββββββββββββββββββ
st.markdown('π€ Agent Execution
', unsafe_allow_html=True)
# Status placeholders
status_ph = {}
for key, meta in agent_meta.items():
status_ph[key] = st.empty()
status_ph[key].markdown(
f''
f''
f'{meta["icon"]} {meta["label"]} β Waitingβ¦
',
unsafe_allow_html=True,
)
# Progress bar (5 nodes: 4 agents + 1 scoring = 20% each)
progress_bar = st.progress(0, text="Starting agentsβ¦")
initial_state: AgentState = {
"idea": idea,
"results": [],
"score": 0,
"verdict": "",
"summary": "",
}
start_time = time.time()
final_state = None
all_results = []
nodes_done = 0
for chunk in app.stream(initial_state, stream_mode="updates"):
for node_name, state_update in chunk.items():
nodes_done += 1
pct = min(nodes_done / 5, 1.0)
if node_name in status_ph:
meta = agent_meta[node_name]
status_ph[node_name].markdown(
f''
f''
f'β
{meta["icon"]} {meta["label"]} β Complete
',
unsafe_allow_html=True,
)
# Update progress bar
if pct < 1.0:
progress_bar.progress(pct, text=f"Agents working⦠{int(pct * 100)}%")
else:
progress_bar.progress(1.0, text="All agents finished!")
# Collect results
if "results" in state_update:
all_results.extend(state_update["results"])
if node_name == "scoring_node":
final_state = state_update
elapsed = time.time() - start_time
# Cache the results
st.session_state.cache[idea_key] = {
"final_state": final_state,
"all_results": all_results,
"elapsed": elapsed,
}
st.markdown(
f''
f'β¨ Validated in {elapsed:.1f}s
',
unsafe_allow_html=True,
)
st.markdown('', unsafe_allow_html=True)
# ββββββββββββββββββββββββββββββββββββββ
# VERDICT + SUMMARY (shared path)
# ββββββββββββββββββββββββββββββββββββββ
if final_state:
score = final_state["score"]
verdict = final_state["verdict"]
summary = final_state["summary"]
# Style by verdict
if verdict == "Go":
badge_cls = "badge-go"
score_grad = "linear-gradient(135deg, #059669, #34d399)"
elif verdict == "Caution":
badge_cls = "badge-caution"
score_grad = "linear-gradient(135deg, #d97706, #fbbf24)"
else:
badge_cls = "badge-nogo"
score_grad = "linear-gradient(135deg, #dc2626, #f87171)"
col_v, col_s = st.columns([1, 2], gap="large")
with col_v:
st.markdown(f"""
{score}
confidence score
{verdict}
""", unsafe_allow_html=True)
with col_s:
st.markdown(f"""
π Executive Summary
{summary}
""", unsafe_allow_html=True)
st.markdown('', unsafe_allow_html=True)
# ββββββββββββββββββββββββββββββββββ
# INDIVIDUAL AGENT REPORTS
# ββββββββββββββββββββββββββββββββββ
st.markdown('π Detailed Agent Reports
', unsafe_allow_html=True)
report_style = {
"competitor_analysis": {"cls": "rc-competitor", "icon_cls": "rc-icon-comp", "icon": "π", "label": "Competitor Analysis"},
"market_sizing": {"cls": "rc-market", "icon_cls": "rc-icon-mkt", "icon": "π", "label": "Market Sizing"},
"community_sentiment": {"cls": "rc-sentiment", "icon_cls": "rc-icon-sent", "icon": "π¬", "label": "Community Sentiment"},
"os_risk": {"cls": "rc-osrisk", "icon_cls": "rc-icon-osr", "icon": "β οΈ", "label": "Open-Source Risk"},
}
# 2 Γ 2 grid
left_col, right_col = st.columns(2, gap="medium")
for idx, result in enumerate(all_results):
agent_key = result.get("agent", "unknown")
style = report_style.get(agent_key, {"cls": "rc-competitor", "icon_cls": "rc-icon-comp", "icon": "π", "label": agent_key})
findings = result.get("findings", "No findings available.")
sources = result.get("sources", [])
# Escape any stray HTML in LLM output but keep line breaks
findings_html = findings.replace("&", "&").replace("<", "<").replace(">", ">").replace("\n", "
")
# Convert markdown bold/italic to HTML
findings_html = re.sub(r'\*\*(.+?)\*\*', r'\1', findings_html)
findings_html = re.sub(r'\*(.+?)\*', r'\1', findings_html)
# Build source pills HTML (ChatGPT-style)
sources_html = ""
if sources:
pills = ""
for i, src in enumerate(sources):
favicon = f"https://www.google.com/s2/favicons?domain={src['domain']}&sz=32"
title = src['title'][:42]
domain = src['domain'][:25]
pills += (
f''
f'
'
f'{i + 1}'
f'{title}'
f'{domain}'
f''
)
sources_html = (
f''
f'
Sources
'
f'
{pills}
'
f'
'
)
card_html = f"""
{findings_html}
{sources_html}
"""
if idx % 2 == 0:
with left_col:
st.markdown(card_html, unsafe_allow_html=True)
else:
with right_col:
st.markdown(card_html, unsafe_allow_html=True)
st.markdown('', unsafe_allow_html=True)
# ββββββββββββββββββββββββββββββββββ
# PDF DOWNLOAD BUTTON
# ββββββββββββββββββββββββββββββββββ
pdf_bytes = generate_pdf(idea, score, verdict, summary, all_results)
timestamp = datetime.now().strftime("%Y%m%d_%H%M")
dl_col1, dl_col2, dl_col3 = st.columns([1, 1, 1])
with dl_col2:
st.download_button(
label="π Download PDF Report",
data=pdf_bytes,
file_name=f"startup_validation_{timestamp}.pdf",
mime="application/pdf",
)
# ββ Footer ββ
st.markdown(
'',
unsafe_allow_html=True,
)