| """ | |
| 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("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 = [] | |
| 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)) | |
| 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) | |
| return | |
| history.append((user_input, f"Analyzing: '{query_context}'\n\n*Generating analysis plan...*")) | |
| yield (history, state, None, None, None, None, None) | |
| analysis_plan = llm_generate_analysis_plan_with_history(llm_model, query_context, history) | |
| 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) | |
| 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) | |
| history.append((None, "*Executing queries for aggregates and examples...*")) | |
| yield (history, state, None, None, gr.update(value=formatted_plan, visible=True), None, None) | |
| 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) | |
| return | |
| 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) | |
| 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) | |
| 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) | |
| 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, report_text, gr.update(value=formatted_plan, visible=True), gr.update(value=formatted_agg_data, visible=True), qual_data_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, report_text, gr.update(value=formatted_plan, visible=True), gr.update(value=formatted_agg_data, visible=True), qual_data_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) | |
| ) | |
| 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], | |
| ).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], | |
| queue=False | |
| ) | |
| return demo | |