+
+
+💡
+
+
Generate Synthetic Data
+
+
Define the scope, constraints, and target context for your synthetic data generation pipeline.
+
+
+
+🕸️
+
+
Refine Taxonomy
+
+
Review the generated taxonomy and associated relationships. Edit branches of the taxonomy or proceed to grounded data generation.
+
+🌍 {regions}
+📌 {concept_name}
+
+
', unsafe_allow_html=True)
+ col_tree, col_meta = st.columns([1, 1], gap="large")
+
+ with col_tree:
+ st.markdown("### L1, L2, L3 Tree")
+ l1_groups = tree_df.groupby('level1')
+ for l1, l1_df in l1_groups:
+ with st.expander(f"📁 **L1** {l1}", expanded=True):
+ l2_groups = l1_df.groupby('level2')
+ for l2, l2_df in l2_groups:
+ with st.expander(f"📂 **L2** {l2}", expanded=True):
+ l3_items = sorted(l2_df['level3'].unique())
+ for l3 in l3_items:
+ if st.button(f"👉 **L3** {l3}", key=f"btn_{l1}_{l2}_{l3}", use_container_width=True):
+ st.session_state.selected_l3 = l3
+
+ # CSS specifically for the L3 buttons to make them look clickable
+ st.markdown("""
+
+ """, unsafe_allow_html=True)
+
+ with col_meta:
+ st.markdown("### Metadata Details")
+ st.markdown("
", unsafe_allow_html=True)
+
+ if 'selected_l3' not in st.session_state or not st.session_state.selected_l3:
+ st.info("ℹ️ Metadata is available for L3 Leaf nodes. Click any L3 node in the tree to view its details.")
+ else:
+ # Find data for the selected L3
+ # We search in the exploded demo_data to find matching row
+ df_search = df.copy()
+ df_search['level3_list'] = df_search['level3'].apply(safe_eval_list)
+ df_exploded = df_search.explode('level3_list')
+ match = df_exploded[df_exploded['level3_list'] == st.session_state.selected_l3]
+
+ if not match.empty:
+ node_data = match.iloc[0]
+
+ st.markdown(f"#### {st.session_state.selected_l3}")
+
+ # Geographic Context
+ st.markdown("🌐 **GEOGRAPHIC CONTEXT**")
+ # Prefer extracted_Country, fallback to cleaned_Country, fallback to default
+ country_val = node_data.get('extracted_Country', node_data.get('cleaned_Country', 'Global'))
+ # Format as blue pill if it's string, handle lists
+ if isinstance(country_val, str) and country_val.startswith('['):
+ try: country_val = eval(country_val)
+ except: pass
+ if isinstance(country_val, list):
+ pills = "".join([f"
{c}" for c in country_val])
+ st.markdown(pills, unsafe_allow_html=True)
+ else:
+ st.markdown(f"
{country_val}", unsafe_allow_html=True)
+
+ st.markdown("
", unsafe_allow_html=True)
+
+ # Demographics
+ st.markdown("👥 **DEMOGRAPHICS**")
+ # Use extracted_Demographics or user_group
+ demo_val = node_data.get('extracted_Demographics', node_data.get('user_group', 'N/A'))
+ if isinstance(demo_val, str) and demo_val.startswith('['):
+ try: demo_val = eval(demo_val)
+ except: pass
+ if isinstance(demo_val, list):
+ pills = "".join([f"
{d}" for d in demo_val])
+ st.markdown(pills, unsafe_allow_html=True)
+ else:
+ st.markdown(f"
{demo_val}", unsafe_allow_html=True)
+
+ st.markdown("
", unsafe_allow_html=True)
+
+ # Use Cases
+ st.markdown("📋 **USE CASES**")
+ use_case = node_data.get('user_case', 'N/A')
+ st.markdown(f"
{use_case}
", unsafe_allow_html=True)
+
+ st.markdown("
", unsafe_allow_html=True)
+
+ # Research Citations
+ st.markdown("📖 **RESEARCH CITATIONS**")
+
+ title = "Research Paper"
+ import re
+ paper = str(node_data.get('paper_content', ''))
+ title_match = re.search(r'\*\*Title:\*\*\s*(?:\"(.*?)\"|(.*?)\n)', paper)
+ if title_match:
+ title = (title_match.group(1) or title_match.group(2) or "Citation 1").strip()
+
+ url_val = node_data.get('url', '')
+ if isinstance(url_val, str) and url_val.startswith('['):
+ try: url_val = eval(url_val)
+ except: pass
+
+ if isinstance(url_val, list):
+ if url_val:
+ st.markdown(f"- [{title}]({url_val[0]})")
+ elif isinstance(url_val, str) and url_val:
+ st.markdown(f"- [{title}]({url_val})")
+ else:
+ # fallback paper content
+ paper = node_data.get('paper_content', '')
+ if paper:
+ st.info("Citations available in internal knowledge source.")
+ else:
+ st.write("No citations available.")
+ else:
+ st.warning("Data not found for this node.")
+ st.markdown('
+
+
+🗄️
+
+
Review Synthetic Data
+
+
Assess the quality and diversity of the grounded synthetic data you created.
+
+🌍 {regions}
+📌 {concept_name}
+
+
+
Evaluate Your Target Model
+
Generate responses from the model you want to evaluate.
+
+
+
+📝
+
+
Define Autorator
+
+
Configure the evaluation rubric and provide structured feedback on model annotations.
+
+
+
+📊
+
+
Error Analysis Dashboard
+
+
Disclosure compliance analysis across models and data sources. Identify failure patterns and missing medical disclaimers.
+
No bi-grams shared across all models.
',
+ unsafe_allow_html=True)
+
+ unique_cols = st.columns(len(models))
+ for idx, m in enumerate(models):
+ with unique_cols[idx]:
+ m_color = MODEL_COLORS.get(m, default_color)["line"]
+ unique_list = sorted(unique_phrases.get(m, set()))
+ st.markdown(f"""
+