File size: 10,513 Bytes
f43f2d3 f74c067 f43f2d3 f74c067 f43f2d3 f74c067 f43f2d3 f74c067 f43f2d3 f74c067 f43f2d3 f74c067 f43f2d3 f74c067 f43f2d3 f74c067 f43f2d3 f74c067 f43f2d3 f74c067 f43f2d3 f74c067 f43f2d3 f74c067 f43f2d3 f74c067 f43f2d3 f74c067 f43f2d3 f74c067 f43f2d3 f74c067 f43f2d3 f74c067 f43f2d3 f74c067 f43f2d3 f74c067 f43f2d3 f74c067 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 |
"""
Defines the Gradio user interface and manages the application's state
and event handling.
This module is responsible for the presentation layer of the application.
It creates the interactive components and orchestrates the analysis workflow
by calling functions from the data_processing module.
"""
import gradio as gr
import json
import concurrent.futures
from data_processing import (
llm_generate_analysis_plan_with_history,
execute_quantitative_query,
execute_qualitative_query,
llm_synthesize_enriched_report_stream,
llm_generate_visualization_code,
execute_viz_code_and_get_path,
parse_suggestions_from_report
)
def create_ui(llm_model, solr_client):
"""
Builds the Gradio UI and wires up all the event handlers.
Args:
llm_model: The initialized Google Gemini model client.
solr_client: The initialized pysolr client.
"""
with gr.Blocks(theme=gr.themes.Soft(), css="footer {display: none !important}") as demo:
state = gr.State()
with gr.Row():
with gr.Column(scale=4):
gr.Markdown("# PharmaCircle AI Data Analyst")
with gr.Column(scale=1):
clear_button = gr.Button("🔄 Start New Analysis", variant="primary")
gr.Markdown("Ask a question to begin your analysis. I will generate an analysis plan, retrieve quantitative and qualitative data, create a visualization, and write an enriched report.")
with gr.Row():
with gr.Column(scale=1):
chatbot = gr.Chatbot(label="Analysis Chat Log", height=700, show_copy_button=True)
msg_textbox = gr.Textbox(placeholder="Ask a question, e.g., 'Show me the top 5 companies by total deal value in 2023'", label="Your Question", interactive=True)
with gr.Column(scale=2):
with gr.Accordion("Dynamic Field Suggestions", open=False):
suggestions_display = gr.Markdown("Suggestions from the external API will appear here...", visible=True)
with gr.Accordion("Generated Analysis Plan", open=False):
plan_display = gr.Markdown("Plan will appear here...", visible=True)
with gr.Accordion("Retrieved Quantitative Data", open=False):
quantitative_data_display = gr.Markdown("Aggregate data will appear here...", visible=False)
with gr.Accordion("Retrieved Qualitative Data (Examples)", open=False):
qualitative_data_display = gr.Markdown("Example data will appear here...", visible=False)
plot_display = gr.Image(label="Visualization", type="filepath", visible=False)
report_display = gr.Markdown("Report will be streamed here...", visible=False)
def process_analysis_flow(user_input, history, state):
"""
Manages the conversation and yields UI updates.
"""
if state is None:
state = {'query_count': 0, 'last_suggestions': []}
if history is None:
history = []
# Reset all displays at the beginning of a new flow
yield (history, state, gr.update(value=None, visible=False), gr.update(value=None, visible=False), gr.update(value=None, visible=False), gr.update(value=None, visible=False), gr.update(value=None, visible=False), gr.update(value="Suggestions from the external API will appear here...", visible=False))
query_context = user_input.strip()
if not query_context:
history.append((user_input, "Please enter a question to analyze."))
yield (history, state, None, None, None, None, None, None)
return
history.append((user_input, f"Analyzing: '{query_context}'\n\n*Generating analysis plan...*"))
yield (history, state, None, None, None, None, None, None)
# Generate plan and get search field suggestions
analysis_plan, search_fields = llm_generate_analysis_plan_with_history(llm_model, query_context, history)
# Update and display search field suggestions in its own accordion
if search_fields:
suggestions_md = "**External API Suggestions:**\n" + "\n".join([f"- `{field['field_name']}`: `{field['field_value']}`" for field in search_fields])
suggestions_display_update = gr.update(value=suggestions_md, visible=True)
else:
suggestions_display_update = gr.update(value="No suggestions were returned from the external API.", visible=True)
if not analysis_plan:
history.append((None, "I'm sorry, I couldn't generate a valid analysis plan. Please try rephrasing."))
yield (history, state, None, None, None, None, None, suggestions_display_update)
return
history.append((None, "✅ Analysis plan generated!"))
plan_summary = f"""
* **Analysis Dimension:** `{analysis_plan.get('analysis_dimension')}`
* **Analysis Measure:** `{analysis_plan.get('analysis_measure')}`
* **Query Filter:** `{analysis_plan.get('query_filter')}`
"""
history.append((None, plan_summary))
formatted_plan = f"**Full Analysis Plan:**\n```json\n{json.dumps(analysis_plan, indent=2)}\n```"
yield (history, state, None, None, gr.update(value=formatted_plan, visible=True), None, None, suggestions_display_update)
history.append((None, "*Executing queries for aggregates and examples...*"))
yield (history, state, None, None, gr.update(value=formatted_plan, visible=True), None, None, suggestions_display_update)
# Execute queries in parallel
aggregate_data = None
example_data = None
with concurrent.futures.ThreadPoolExecutor() as executor:
future_agg = executor.submit(execute_quantitative_query, solr_client, analysis_plan)
future_ex = executor.submit(execute_qualitative_query, solr_client, analysis_plan)
aggregate_data = future_agg.result()
example_data = future_ex.result()
if not aggregate_data or aggregate_data.get('count', 0) == 0:
history.append((None, "No data was found for your query. Please try a different question."))
yield (history, state, None, None, gr.update(value=formatted_plan, visible=True), None, None, suggestions_display_update)
return
# Display retrieved data
formatted_agg_data = f"**Quantitative (Aggregate) Data:**\n```json\n{json.dumps(aggregate_data, indent=2)}\n```"
formatted_qual_data = f"**Qualitative (Example) Data:**\n```json\n{json.dumps(example_data, indent=2)}\n```"
qual_data_display_update = gr.update(value=formatted_qual_data, visible=True)
yield (history, state, None, None, gr.update(value=formatted_plan, visible=True), gr.update(value=formatted_agg_data, visible=True), qual_data_display_update, suggestions_display_update)
history.append((None, "✅ Data retrieved. Generating visualization and final report..."))
yield (history, state, None, None, gr.update(value=formatted_plan, visible=True), gr.update(value=formatted_agg_data, visible=True), qual_data_display_update, suggestions_display_update)
# Generate viz and report
with concurrent.futures.ThreadPoolExecutor() as executor:
viz_future = executor.submit(llm_generate_visualization_code, llm_model, query_context, aggregate_data)
report_text = ""
stream_history = history[:]
for chunk in llm_synthesize_enriched_report_stream(llm_model, query_context, aggregate_data, example_data, analysis_plan):
report_text += chunk
yield (stream_history, state, None, gr.update(value=report_text, visible=True), gr.update(value=formatted_plan, visible=True), gr.update(value=formatted_agg_data, visible=True), qual_data_display_update, suggestions_display_update)
history.append((None, report_text))
viz_code = viz_future.result()
plot_path = execute_viz_code_and_get_path(viz_code, aggregate_data)
output_plot = gr.update(value=plot_path, visible=True) if plot_path else gr.update(visible=False)
if not plot_path:
history.append((None, "*I was unable to generate a plot for this data.*\n"))
yield (history, state, output_plot, gr.update(value=report_text), gr.update(value=formatted_plan, visible=True), gr.update(value=formatted_agg_data, visible=True), qual_data_display_update, suggestions_display_update)
state['query_count'] += 1
state['last_suggestions'] = parse_suggestions_from_report(report_text)
next_prompt = "Analysis complete. What would you like to explore next?"
history.append((None, next_prompt))
yield (history, state, output_plot, gr.update(value=report_text), gr.update(value=formatted_plan, visible=True), gr.update(value=formatted_agg_data, visible=True), qual_data_display_update, suggestions_display_update)
def reset_all():
"""Resets the entire UI for a new analysis session."""
return (
[],
None,
"",
gr.update(value=None, visible=False),
gr.update(value=None, visible=False),
gr.update(value=None, visible=False),
gr.update(value=None, visible=False),
gr.update(value=None, visible=False),
gr.update(value=None, visible=False)
)
msg_textbox.submit(
fn=process_analysis_flow,
inputs=[msg_textbox, chatbot, state],
outputs=[chatbot, state, plot_display, report_display, plan_display, quantitative_data_display, qualitative_data_display, suggestions_display],
).then(
lambda: gr.update(value=""),
None,
[msg_textbox],
queue=False,
)
clear_button.click(
fn=reset_all,
inputs=None,
outputs=[chatbot, state, msg_textbox, plot_display, report_display, plan_display, quantitative_data_display, qualitative_data_display, suggestions_display],
queue=False
)
return demo |