Enhance Reaction SVG tab: AI processing happens FIRST before rendering - DeepSeek analyzes and completes reactions before RDKit SVG generation, ensuring all reactions are properly enhanced/validated/completed

app.py

@@ -222,8 +222,8 @@ def fetch_specific_db(db_name, formula):
 
         # Generic table extraction
         tables = soup.find_all('table')
-        df = None
-        if tables:
+    df = None
+    if tables:
             df = pd.read_html(StringIO(str(tables[0])))[0]
 
         # Fallback: Links for spectra/structures
@@ -900,7 +900,7 @@ def render_reaction_svg(reaction_text: str, api_key: str = "", auto_complete: bo
 
     # Check if it's already SMILES format (contains >>)
     if ">>" in reaction_text:
-        svg = _render_smiles_to_svg(reaction_text)
+    svg = _render_smiles_to_svg(reaction_text)
         if svg:
             status = "✅ Reaction rendered successfully from SMILES."
             return svg, status
@@ -917,7 +917,7 @@ def render_reaction_svg(reaction_text: str, api_key: str = "", auto_complete: bo
                 svg = _render_reaction_from_nist(completed_reaction)
                 if svg:
                     status = f"✅ Reaction completed and rendered using DeepSeek AI.\nOriginal: {reaction_text}\nCompleted: {completed_reaction}"
-                    return svg, status
+    return svg, status
                 else:
                     return "", f"🚨 DeepSeek completed reaction but rendering failed: {completed_reaction}"
             else:
@@ -1065,13 +1065,13 @@ def fetch_detail(selected_url: str, manual_url: str, auto_fetch_thermo: bool = T
         return markdown, table, None, "", {}, ""
 
     plot_fig = _build_dataset_plot(detail)
-
+    
     # Try to render the reaction title as SVG
     reaction_svg = ""
     if detail.title:
         title = detail.title.strip()
         smiles_attempt = None
-
+        
         # Try different reaction format conversions
         if " → " in title:
             # Format: "A + B → C"
@@ -1090,12 +1090,12 @@ def fetch_detail(selected_url: str, manual_url: str, auto_fetch_thermo: bool = T
                 reactants = parts[0].replace(" + ", ".").strip()
                 products = parts[1].replace(" + ", ".").strip()
                 smiles_attempt = f"{reactants}>>{products}"
-
+        
         if smiles_attempt:
             svg = _render_smiles_to_svg(smiles_attempt)
             if svg:
                 reaction_svg = svg
-
+    
     # Auto-fetch thermodynamic data for compounds in the reaction
     thermo_data = {}
     thermo_summary = ""
@@ -1265,24 +1265,24 @@ def build_interface() -> gr.Blocks:
                             value=True,
                             info="Automatically fetch thermodynamic data for searched compounds"
                         )
-
-                with gr.Row():
-                    decomp = gr.Checkbox(label="Only decomposition reactions", value=False)
-                    category = gr.Dropdown(label="Result type filter", choices=CATEGORY_CHOICES, value=str(Category.any.value))
-                    units = gr.Textbox(
-                        label="Optional Units token",
-                        placeholder="Leave blank to use NIST account defaults",
-                    )
+            
+            with gr.Row():
+                decomp = gr.Checkbox(label="Only decomposition reactions", value=False)
+                category = gr.Dropdown(label="Result type filter", choices=CATEGORY_CHOICES, value=str(Category.any.value))
+                units = gr.Textbox(
+                    label="Optional Units token",
+                    placeholder="Leave blank to use NIST account defaults",
+                )
 
                 search_button = gr.Button("🔍 Search NIST", variant="primary")
-                search_status = gr.Markdown()
+            search_status = gr.Markdown()
 
-                result_table = gr.Dataframe(
-                    headers=["#", "Records", "Reaction", "Detail URL"],
-                    datatype=["number", "number", "str", "str"],
-                    interactive=False,
-                    wrap=True,
-                )
+            result_table = gr.Dataframe(
+                headers=["#", "Records", "Reaction", "Detail URL"],
+                datatype=["number", "number", "str", "str"],
+                interactive=False,
+                wrap=True,
+            )
 
                 # Search results thermodynamic data
                 search_thermo_accordion = gr.Accordion(label="🔬 Search Query Thermodynamic Data", open=False)
@@ -1293,15 +1293,15 @@ def build_interface() -> gr.Blocks:
             with gr.TabItem("Reaction Detail"):
                 with gr.Row():
                     with gr.Column(scale=2):
-                        selection = gr.Dropdown(
-                            label="Select a reaction from the latest search",
-                            choices=[],
-                            interactive=False,
-                        )
-                        manual_url = gr.Textbox(
-                            label="Or paste a NIST detail URL",
-                            placeholder="https://kinetics.nist.gov/kinetics/ReactionSearch?....",
-                        )
+            selection = gr.Dropdown(
+                label="Select a reaction from the latest search",
+                choices=[],
+                interactive=False,
+            )
+            manual_url = gr.Textbox(
+                label="Or paste a NIST detail URL",
+                placeholder="https://kinetics.nist.gov/kinetics/ReactionSearch?....",
+            )
                     with gr.Column(scale=1):
                         auto_fetch_thermo = gr.Checkbox(
                             label="🔬 Auto-fetch thermodynamics",
@@ -1315,31 +1315,31 @@ def build_interface() -> gr.Blocks:
                         )
 
                 detail_button = gr.Button("Fetch Reaction Detail", variant="primary")
-
-                # Reaction metadata and details
-                detail_markdown = gr.Markdown()
-
+            
+            # Reaction metadata and details
+            detail_markdown = gr.Markdown()
+            
                 with gr.Row():
-                    # Kinetics data table
+            # Kinetics data table
                     with gr.Column():
                         gr.Markdown("### Kinetics Data")
-                        dataset_table = gr.Dataframe(
-                            headers=["Section", "Squib", "Temp [K]", "A", "n", "Ea [J/mole]", "k(298 K)", "Order", "Squib URL"],
-                            datatype=["str"] * 9,
-                            interactive=False,
-                            wrap=True,
-                        )
-
+            dataset_table = gr.Dataframe(
+                headers=["Section", "Squib", "Temp [K]", "A", "n", "Ea [J/mole]", "k(298 K)", "Order", "Squib URL"],
+                datatype=["str"] * 9,
+                interactive=False,
+                wrap=True,
+            )
+            
                     # Arrhenius plot
                     with gr.Column():
                         gr.Markdown("### Arrhenius Plot")
                         reaction_plot = gr.Plot()
 
-                # Reaction SVG visualization
-                with gr.Row():
-                    gr.Markdown("### Reaction Structure")
-                reaction_svg = gr.HTML()
-
+            # Reaction SVG visualization
+            with gr.Row():
+                gr.Markdown("### Reaction Structure")
+            reaction_svg = gr.HTML()
+            
                 # Auto-fetched thermodynamic data
                 thermo_summary = gr.Markdown()
                 thermo_accordion = gr.Accordion(label="🔬 Thermodynamic Data", open=False)
@@ -1350,13 +1350,18 @@ def build_interface() -> gr.Blocks:
             # Tab 3: Reaction SVG (Enhanced with NIST reactions and AI completion)
             with gr.TabItem("Reaction SVG"):
                 gr.Markdown(
-                    "🎨 **Render chemical reactions as SVG using RDKit + AI Completion**\n\n"
-                    "Choose from NIST database reactions, enter custom reactions, or let AI complete partial reactions!\n\n"
+                    "🎨 **Render chemical reactions as SVG using RDKit + AI Enhancement**\n\n"
+                    "Choose from NIST database reactions, enter custom reactions, or let AI enhance/validate/complete your reactions!\n\n"
+                    "**Workflow:**\n"
+                    "1. 🤖 **AI Enhancement First**: DeepSeek AI analyzes and enhances your reaction\n"
+                    "2. 🎨 **RDKit Rendering**: Complete reaction rendered as beautiful SVG\n"
+                    "3. ✅ **Validation**: AI confirms reaction balance and plausibility\n\n"
                     "**Features:**\n"
                     "- 🧪 200+ NIST database reactions\n"
-                    "- 🤖 AI-powered reaction completion (DeepSeek)\n"
-                    "- 🔬 Multiple input formats (NIST, SMILES, SMARTS)\n"
-                    "- ⚡ Automatic format detection and conversion"
+                    "- 🤖 AI-powered reaction enhancement (DeepSeek-V3.2-Exp)\n"
+                    "- 🔬 Multiple input formats (NIST, SMILES, SMARTS, partial)\n"
+                    "- ⚡ Automatic format detection and intelligent conversion\n"
+                    "- ✅ Reaction validation and balancing"
                 )
 
                 # API Key Configuration
@@ -1367,7 +1372,7 @@ def build_interface() -> gr.Blocks:
                         type="password",
                         info="Get your API key from https://platform.deepseek.com/"
                     )
-                    gr.Markdown(
+            gr.Markdown(
                         "**How to get API key:**\n"
                         "1. Visit https://platform.deepseek.com/\n"
                         "2. Sign up/Login to your account\n"
@@ -1379,8 +1384,8 @@ def build_interface() -> gr.Blocks:
                 # NIST reactions dropdown
                 nist_reactions = _fetch_all_nist_reactions(limit=200)
                 nist_reaction_options = [("", "")] + nist_reactions if nist_reactions else []
-
-                with gr.Row():
+            
+            with gr.Row():
                     with gr.Column():
                         nist_reaction_dropdown = gr.Dropdown(
                             label="🧪 NIST Database Reactions",
@@ -1389,8 +1394,8 @@ def build_interface() -> gr.Blocks:
                             interactive=True,
                             info=f"Select from {len(nist_reactions)} reactions in NIST kinetics database"
                         )
-
-                        reaction_input = gr.Textbox(
+            
+            reaction_input = gr.Textbox(
                             label="Custom Reaction Input",
                             placeholder="Enter reaction in any format:\nNIST: CH4 + O2 → CO2 + H2O\nSMILES: CH4.O2>>CO2.H2O\nPartial: CH4 + O2 → (AI will complete)",
                             lines=4,
@@ -1406,17 +1411,20 @@ def build_interface() -> gr.Blocks:
                         )
 
                         ai_options = gr.CheckboxGroup(
-                            label="🤖 AI Enhancement",
-                            choices=["Enable AI completion", "High quality rendering", "Show completion reasoning"],
-                            value=["Enable AI completion"],
-                            info="Use DeepSeek AI to complete partial reactions"
+                            label="🤖 AI Enhancement Options",
+                            choices=["Enable AI enhancement (recommended)", "High quality rendering", "Show AI reasoning"],
+                            value=["Enable AI enhancement (recommended)"],
+                            info="DeepSeek AI analyzes, validates, and enhances reactions before rendering"
                         )
 
                 # Buttons
                 with gr.Row():
-                    render_auto_btn = gr.Button("🚀 Smart Render (Auto-detect)", variant="primary")
-                    render_nist_btn = gr.Button("🧪 Render NIST Format", variant="secondary")
-                    render_smiles_btn = gr.Button("🔬 Render SMILES/SMARTS", variant="secondary")
+                    render_auto_btn = gr.Button("🚀 AI First → Render (Recommended)", variant="primary",
+                                               info="DeepSeek AI enhances/completes reaction first, then renders SVG")
+                    render_nist_btn = gr.Button("🧪 Direct NIST Render", variant="secondary",
+                                               info="Render directly without AI enhancement")
+                    render_smiles_btn = gr.Button("🔬 Direct SMILES Render", variant="secondary",
+                                                 info="Render SMILES/SMARTS directly without AI enhancement")
                     clear_btn = gr.Button("🗑️ Clear", variant="stop")
 
                 # Output
@@ -1428,14 +1436,14 @@ def build_interface() -> gr.Blocks:
                 def populate_from_nist_dropdown(selected_label):
                     if selected_label and selected_label in nist_dict:
                         return nist_dict[selected_label]
-                    return ""
-
+                return ""
+            
                 nist_reaction_dropdown.change(
                     fn=populate_from_nist_dropdown,
                     inputs=nist_reaction_dropdown,
-                    outputs=reaction_input,
-                )
-
+                outputs=reaction_input,
+            )
+            
                 # Smart auto-render function
                 def render_auto_reaction(reaction_text, api_key, ai_options, render_mode):
                     if not reaction_text:
@@ -1444,35 +1452,52 @@ def build_interface() -> gr.Blocks:
                     status_prefix = ""
                     final_reaction = reaction_text
 
-                    # Check if AI completion is enabled and reaction might be incomplete
-                    if "Enable AI completion" in (ai_options or []) and api_key:
-                        analysis = _analyze_reaction_completeness(reaction_text)
-                        if not analysis["complete"]:
-                            completed_reaction = _complete_reaction_with_deepseek(reaction_text, api_key)
-                            if completed_reaction:
-                                final_reaction = completed_reaction
-                                status_prefix = f"🤖 **AI Completed Reaction**\nOriginal: {reaction_text}\nCompleted: {final_reaction}\n\n"
-
-                    # Render based on detected format or forced mode
-                    if render_mode == "Force SMILES/SMARTS" or (render_mode == "Auto (detect format)" and ">>" in final_reaction):
+                    # Always try AI enhancement first if enabled and API key provided
+                    if "Enable AI enhancement (recommended)" in (ai_options or []) and api_key:
+                        # Try to complete/enhance the reaction using DeepSeek
+                        completed_reaction = _complete_reaction_with_deepseek(reaction_text, api_key)
+                        if completed_reaction and completed_reaction != reaction_text:
+                            final_reaction = completed_reaction
+                            status_prefix = f"🤖 **AI Enhanced Reaction**\nOriginal: {reaction_text}\nAI Completed: {final_reaction}\n\n"
+                        elif completed_reaction == reaction_text:
+                            # AI validated the reaction as complete
+                            status_prefix = f"🤖 **AI Validated Reaction**\nReaction confirmed as complete and balanced.\n\n"
+                        else:
+                            # AI failed, try direct rendering
+                            status_prefix = f"⚠️ **AI Enhancement Failed**\nProceeding with original reaction.\n\n"
+
+                    # Render the final reaction (AI-enhanced or original)
+                    svg = None
+                    render_type = "unknown"
+
+                    # Try different rendering approaches based on mode
+                    if render_mode == "Force SMILES/SMARTS":
                         svg = _render_smiles_to_svg(final_reaction)
                         render_type = "SMILES/SMARTS"
-                    elif render_mode == "Force NIST format" or render_mode == "Auto (detect format)":
+                    elif render_mode == "Force NIST format":
                         svg = _render_reaction_from_nist(final_reaction)
                         render_type = "NIST format"
-                    else:
-                        # Try both formats
-                        svg = _render_reaction_from_nist(final_reaction)
-                        if not svg:
+                    else:  # Auto (detect format)
+                        # First try SMILES if it contains >>
+                        if ">>" in final_reaction:
                             svg = _render_smiles_to_svg(final_reaction)
-                        render_type = "auto-detected"
+                            render_type = "SMILES/SMARTS (detected)"
+                        else:
+                            # Try NIST format first, then SMILES
+                            svg = _render_reaction_from_nist(final_reaction)
+                            if svg:
+                                render_type = "NIST format (detected)"
+                            else:
+                                svg = _render_smiles_to_svg(final_reaction)
+                                render_type = "SMILES/SMARTS (fallback)"
 
                     if svg:
                         quality_note = " (High quality)" if "High quality rendering" in (ai_options or []) else ""
-                        status = f"{status_prefix}✅ Successfully rendered as {render_type}{quality_note}"
+                        reasoning_note = " (with AI reasoning)" if "Show AI reasoning" in (ai_options or []) else ""
+                        status = f"{status_prefix}✅ Successfully rendered as {render_type}{quality_note}{reasoning_note}"
                         return svg, status
                     else:
-                        return "", f"{status_prefix}❌ Could not render reaction. Try a different format or check syntax: {final_reaction[:100]}..."
+                        return "", f"{status_prefix}❌ Could not render reaction. The reaction format may not be supported: {final_reaction[:100]}...\n\nTry adjusting the render mode or checking your reaction syntax."
 
                 # Legacy render functions (kept for compatibility)
                 def render_nist_reaction(reaction_text, options):
@@ -1482,7 +1507,7 @@ def build_interface() -> gr.Blocks:
                     svg = _render_reaction_from_nist(reaction_text)
                     if svg:
                         status = f"✅ Successfully rendered NIST reaction: {reaction_text[:100]}..."
-                        if "High quality" in (options or []):
+                        if "High quality rendering" in (options or []):
                             status += " (High quality mode)"
                         return svg, status
                     else:
@@ -1495,7 +1520,7 @@ def build_interface() -> gr.Blocks:
                     svg = _render_smiles_to_svg(reaction_text)
                     if svg:
                         status = f"✅ Successfully rendered SMILES reaction: {reaction_text[:100]}..."
-                        if "High quality" in (options or []):
+                        if "High quality rendering" in (options or []):
                             status += " (High quality mode)"
                         return svg, status
                     else:
@@ -1509,8 +1534,8 @@ def build_interface() -> gr.Blocks:
                 render_auto_btn.click(
                     fn=render_auto_reaction,
                     inputs=[reaction_input, deepseek_api_key, ai_options, render_mode],
-                    outputs=[reaction_svg_output, render_status],
-                )
+                outputs=[reaction_svg_output, render_status],
+            )
 
                 render_nist_btn.click(
                     fn=render_nist_reaction,
@@ -1532,8 +1557,8 @@ def build_interface() -> gr.Blocks:
 
             # Tab 4: Kinetics Plotter
             with gr.TabItem("Kinetics Plotter"):
-                with gr.Row():
-                    with gr.Column():
+                        with gr.Row():
+                            with gr.Column():
                         A_input = gr.Number(value=1.3e-9, label="A (cm³/molecule·s)")
                         n_input = gr.Number(value=-0.495, label="n (power)")
                         Ea_input = gr.Number(value=1150, label="Ea (J/mol)")
@@ -1571,8 +1596,8 @@ def build_interface() -> gr.Blocks:
 
                             fetch_btn = gr.Button("Fetch Data")
 
-                        with gr.Column():
-                            output_md = gr.Markdown()
+                            with gr.Column():
+                                output_md = gr.Markdown()
                             output_df = gr.Dataframe(label="Tabular Data")
                             output_plot = gr.Plot(visible=False, label="Spectrum Preview")  # For IR/UV/THz later
 
@@ -1613,7 +1638,7 @@ def build_interface() -> gr.Blocks:
                 ["CH3"], ["benzene"], ["C6H5OH"]
             ],
             inputs=[formula_input]  # Note: This is placeholder; make per-tab if needed
-        )
+                        )
 
     return demo