Spaces:

10gen
/

deepsearchitv2

Running

App Files Files Community

Guiyom commited on Feb 23, 2025

Commit

bfb85f0

verified ·

1 Parent(s): c771d4a

Update app.py

Browse files

Files changed (1) hide show

app.py +28 -25

app.py CHANGED Viewed

@@ -41,7 +41,7 @@ The output MUST include only:
  • The graph code (using D3.js) that uses data relevant to:
 {placeholder_text}
 Ensure that no extraneous content (such as discussion, concluding remarks, or additional text) is included.
-The resulting iframe content should render at full page width and be no taller than 600px.
 Output the result starting with <html> and ending with </html> with no additional comments.
 // Sources:
@@ -87,7 +87,7 @@ Keep in mind the:
       }
       #chart {
         width: 100%;
-        max-height: 600px;
       }
       .axis-label {
         font-size: 14px;
@@ -216,8 +216,8 @@ Keep in mind the:
       margin: 20px 0;
     }
     #chart {
-      width: 100%;
-      max-height: 600px;
     }
     .axis path,
     .axis line {
@@ -379,8 +379,8 @@ Keep in mind the:
       color: #333;
     }
     #chart {
-      width: 100%;
-      max-height: 600px;
       background-color: #fff;
       border: 1px solid #ddd;
       box-shadow: 0px 0px 5px rgba(0, 0, 0, 0.1);
@@ -525,7 +525,7 @@ Keep in mind the:
     result = result.strip().strip("```").strip()
     htmloutput = f"""<iframe class="visual-frame" srcdoc='
         <!DOCTYPE html>
-{result}' width="600px" height="600px" style="border:none;"></iframe>
         """
     logging.info(f"The code produced for this graph placeholder:\n{placeholder_text}\n\n {htmloutput}\n\n")
@@ -576,7 +576,7 @@ def generate_visual_snippet(placeholder_text: str, context: str, initial_query:
 Generate a mermaid code displaying a simple but effective and elegant visualization based on the following requirements:
 {placeholder_text}
-It will be integrated in a broader report (focus on the mermaid formatting though) about:
 {initial_query}
 // Sources:
@@ -607,7 +607,8 @@ G --> H[Performance Evaluation - 45% Speed Improvement, 35% Risk Profiling, 50%
 Note: Visual placeholders are solely supposed to generate diagrams not for graphs.
 - use only this type of quotation marks: ", others would lead to failure to render
 - if you need to render a quotation mark inside the text of the diagram make sure to add escapechar so that mermaid interprets it correctly
-- do not use commas in the content of each object / nodes, they will create rendering issues
 - Take a deep breath, think step by step and think it well.
 // Examples
@@ -615,9 +616,10 @@ Note: Pay attention for each example to what type of parenthesis / bracket is us
 -- flowchart --
 Important:
 - If the flow is "broader" than deep (>3 branches at the same level), choose LR (Left Right)
-- If the flow is "deeper" than broad (>3 levels), choose TD (Top Down)
 - There can be mix of flows that can be elaborate
 Top Down:
@@ -770,7 +772,7 @@ mindmap
     {result}
                 </div>
             </body>
-        </html>' width="600px" height="600px" style="border:none;"></iframe>
         """
     logging.info(f"The code produced for this visual placeholder:\n{placeholder_text}\n\n {htmloutput}\n\n")
@@ -823,6 +825,7 @@ def generate_focus_snippet(placeholder_text: str, context: str, initial_query: s
 Generate a complete, self-contained inner-HTML code (only the content inside <div> tags, no outer <html> tags) with inline CSS.
 The code should provide an elaborate technical drill‐down on the topic:
 {placeholder_text}
 The output must be between 1000 and 2000 words.
 - Use clear sub‑titles (bold them or use <h3> tags) and paragraphs to break the content.
@@ -837,7 +840,8 @@ Keep in mind the:
 {crumbs}
 // Critical Requirements
-It MUST include inline citations using the format [1], [2], etc., that exactly match the reference numbering used in the overall report. If new sources are introduced in this focus box, mention them with new numbers and ensure they are later added to the reference summary table.
 - Use html with css to render it, keep it sober and focused on the content
 - The output has to be boxed in a border of 1px Black
 - Font size minimum 12px for readability in PDF
@@ -849,7 +853,7 @@ It MUST include inline citations using the format [1], [2], etc., that exactly m
 - Ground the content on the inputs shared above (context and knowledge inputs) with facts and numbers
 - Take a deep breath, think step by step and think it well.
 - no introduction, conclusions, code fences -> Output the result directly
-- title should be h2 level then the content
 """)
     result = openai_call(prompt, model="o3-mini", max_tokens_param=10000)
@@ -1190,7 +1194,7 @@ The report should be very detailed and lengthy — approximately the equivalent
 - Do not add any inline citations reference in the placeholders descriptions below (visual, graph or focus)
 - Don't make too large blocks, skip lines and add line breaks when changing topic.
 - The report must include between {round(pages/3.3,0)} and {round(pages/2,0)} tables from the sources used (add citations if necessary) and use facts and figures extensively to ground the analysis.
-- The report MUST include a reference summary table with at least 20 references. All inline citations (e.g., [1], [2], …) present in the report and in any focus placeholders MUST have a corresponding entry in this table with its full URL.
 - For the numbering of titles or numbered lists, use numbers (ex: 1.) and sub-units (1.1, 1.2... 1.1.1...,1.1.2...).
 Note: Exclude the use of html numbered lists format, they don't get correctly implemented. Use plain text format for numbering of sections and sub-sections
 - For the reference citations, add systematically the urls from the Learnings (no need to put them in numbered list format since we alredy have the [x] that serves as number list)
@@ -1225,17 +1229,17 @@ with:
 - source_name as the main source used and
 - y as the source reference in the reference table.
 Important:
 - after [[ put "Visual Placeholder n:" explicitly (with n as the ref number of the visual box created). This will be used in a regex
-- before [[...]] do not put any title for the placeholder, just mention the source
-- use h4 tag for the source and source reference number
 - the only types of mermaid diagram that can be generated are: flowchart, sequence, gantt, pie, mindmap (no charts) // Take this into consideration when providing the instructions for the diagram
-- do not make reference in the report to "visual placeholders" just mention visuals.
 - in the placeholder, no need to add the references to the source or its ref number, but make sure ALL of the data points required has a source from the learning and reference material hereafter
 - these placeholders text should contain:
     o the purpose of the future visual
     o the relevant data to generate it
 - there should be between {round(pages/3.3,0)} and {round(pages/2.5,0)} of these visuals placeholders within the report (all between introduction and conclusion)
-- 2 visual placeholders cannot be in the same section
 Note: the placeholders will then be processed separately by a llm to generate the specific code to display each of them so the instruction need to be clear enough.
 // Graph placeholders
@@ -1252,9 +1256,9 @@ with:
 - source_name as the main source used and
 - y as the source reference in the reference table.
 Important:
 - after [[ put "Graph Placeholder n:" explicitly (with n as the ref number of the graph created). This will be used in a regex
-- use h4 tag for the source and source reference number
-- before [[...]] do not put any title for the placeholder, just mention the source
 - All types of graphs (using d3.js library) can be generated // Take this into consideration when providing the instructions for the graph data
 - do not make mention in the report to "graph placeholders" just mention graph.
@@ -1269,8 +1273,8 @@ Note: the placeholders will then be processed separately by a llm to generate th
 // Focus placeholders
 - To drill down on specific topic that would be deserve to be developped extensively separately, create special focus placeholders in [[...]] double backets
 Note: outside of the placeholder, do not make reference in the report to "focus placeholders" just mention the "Focus box n".
-- in the placeholder, make a mention to the prescribed sources used
-- do not make the placeholder as the exact same topic as the section or the sub-section where it is positioned, it has to be either:
     o a special case that deserves attention
     o a recent development / innovation
     o a theoretical drill-down
@@ -1281,10 +1285,9 @@ Note: outside of the placeholder, do not make reference in the report to "focus
     o the guidance in terms of style and message to convey
     Note: Be specific if you want some particular point developped, keep it consistent across the report.
 - there should be between {round(pages/5,0)} and {round(pages/3.3,0)} of these focus placeholders within the report (all between introduction and conclusion)
-- 2 focus placeholders cannot be in the same section
 - Mention all the sources that should be used to generate this focus placeholder and list also the references that will be mentioned in the References section later (ex: [1], [2])
-- Tip: Use the placeholders for topics that cannot be developped but could be developped later independently, so that you can focus on others.
-Note: the placeholders will then be processed separately by a llm to generate the specific code to display each of them so the instruction need to be clear enough.
 // Format:
 [[Focus Placeholder n:

  • The graph code (using D3.js) that uses data relevant to:
 {placeholder_text}
 Ensure that no extraneous content (such as discussion, concluding remarks, or additional text) is included.
+Take in consideration that the resulting iframe content will render 1400px width and and 800px high.
 Output the result starting with <html> and ending with </html> with no additional comments.
 // Sources:
       }
       #chart {
         width: 100%;
+        max-height: 800px;
       }
       .axis-label {
         font-size: 14px;
       margin: 20px 0;
     }
     #chart {
+      width: 1400px;
+      max-height: 800px;
     }
     .axis path,
     .axis line {
       color: #333;
     }
     #chart {
+      width: 1400;
+      max-height: 800px;
       background-color: #fff;
       border: 1px solid #ddd;
       box-shadow: 0px 0px 5px rgba(0, 0, 0, 0.1);
     result = result.strip().strip("```").strip()
     htmloutput = f"""<iframe class="visual-frame" srcdoc='
         <!DOCTYPE html>
+{result}' width="1400px" height="800px" style="border:none;"></iframe>
         """
     logging.info(f"The code produced for this graph placeholder:\n{placeholder_text}\n\n {htmloutput}\n\n")
 Generate a mermaid code displaying a simple but effective and elegant visualization based on the following requirements:
 {placeholder_text}
+It will be integrated in a broader report about:
 {initial_query}
 // Sources:
 Note: Visual placeholders are solely supposed to generate diagrams not for graphs.
 - use only this type of quotation marks: ", others would lead to failure to render
 - if you need to render a quotation mark inside the text of the diagram make sure to add escapechar so that mermaid interprets it correctly
+- do not use commas in the content of each object / nodes, they will create rendering issues
+- Take in consideration that the resulting iframe content will render 1400px width and and 800px high.
 - Take a deep breath, think step by step and think it well.
 // Examples
 -- flowchart --
 Important:
+- If the flow is "deeper" than broad (>3 levels), choose TD (Top Down) ==> TD is the default one if you have a doubt
 - If the flow is "broader" than deep (>3 branches at the same level), choose LR (Left Right)
 - There can be mix of flows that can be elaborate
+- Take in consideration that the resulting iframe content will render 1400px width and and 800px high. Make the best of the areas real-estate.
 Top Down:
     {result}
                 </div>
             </body>
+        </html>' width="1400px" height="800px" style="border:none;"></iframe>
         """
     logging.info(f"The code produced for this visual placeholder:\n{placeholder_text}\n\n {htmloutput}\n\n")
 Generate a complete, self-contained inner-HTML code (only the content inside <div> tags, no outer <html> tags) with inline CSS.
 The code should provide an elaborate technical drill‐down on the topic:
 {placeholder_text}
 The output must be between 1000 and 2000 words.
 - Use clear sub‑titles (bold them or use <h3> tags) and paragraphs to break the content.
 {crumbs}
 // Critical Requirements
+- It MUST include inline citations using the format [1], [2], etc., that exactly match the reference numbering used in the overall report.
+Note: no need to add a reference table at the end of the focus box, all the references / citations have already been listed at the end of the report (just refer to the [x] provided in the placeholder instructions)
 - Use html with css to render it, keep it sober and focused on the content
 - The output has to be boxed in a border of 1px Black
 - Font size minimum 12px for readability in PDF
 - Ground the content on the inputs shared above (context and knowledge inputs) with facts and numbers
 - Take a deep breath, think step by step and think it well.
 - no introduction, conclusions, code fences -> Output the result directly
+- title should be h2 level (inside the focus box) then the content
 """)
     result = openai_call(prompt, model="o3-mini", max_tokens_param=10000)
 - Do not add any inline citations reference in the placeholders descriptions below (visual, graph or focus)
 - Don't make too large blocks, skip lines and add line breaks when changing topic.
 - The report must include between {round(pages/3.3,0)} and {round(pages/2,0)} tables from the sources used (add citations if necessary) and use facts and figures extensively to ground the analysis.
+- The report MUST include a reference summary table with between 10 (for a 8 page report) and 30 references (for a 40 pages report). All inline citations (e.g., [1], [2], …) present in the report and in any focus placeholders MUST have a corresponding entry in this table with its full URL.
 - For the numbering of titles or numbered lists, use numbers (ex: 1.) and sub-units (1.1, 1.2... 1.1.1...,1.1.2...).
 Note: Exclude the use of html numbered lists format, they don't get correctly implemented. Use plain text format for numbering of sections and sub-sections
 - For the reference citations, add systematically the urls from the Learnings (no need to put them in numbered list format since we alredy have the [x] that serves as number list)
 - source_name as the main source used and
 - y as the source reference in the reference table.
 Important:
+- before [[...]] do not put any title for the placeholder, just mention the source with the reference number
+- use p tag for the source and source reference number
 - after [[ put "Visual Placeholder n:" explicitly (with n as the ref number of the visual box created). This will be used in a regex
 - the only types of mermaid diagram that can be generated are: flowchart, sequence, gantt, pie, mindmap (no charts) // Take this into consideration when providing the instructions for the diagram
+- do not make reference in the report to "visual placeholders" just mention the visual and the number..
 - in the placeholder, no need to add the references to the source or its ref number, but make sure ALL of the data points required has a source from the learning and reference material hereafter
 - these placeholders text should contain:
     o the purpose of the future visual
     o the relevant data to generate it
 - there should be between {round(pages/3.3,0)} and {round(pages/2.5,0)} of these visuals placeholders within the report (all between introduction and conclusion)
+- 2 visual placeholders cannot be in the same section
 Note: the placeholders will then be processed separately by a llm to generate the specific code to display each of them so the instruction need to be clear enough.
 // Graph placeholders
 - source_name as the main source used and
 - y as the source reference in the reference table.
 Important:
+- before [[...]] do not put any title for the placeholder, just mention the source with the reference number
+- use p tag for the source and source reference number
 - after [[ put "Graph Placeholder n:" explicitly (with n as the ref number of the graph created). This will be used in a regex
 - All types of graphs (using d3.js library) can be generated // Take this into consideration when providing the instructions for the graph data
 - do not make mention in the report to "graph placeholders" just mention graph.
 // Focus placeholders
 - To drill down on specific topic that would be deserve to be developped extensively separately, create special focus placeholders in [[...]] double backets
 Note: outside of the placeholder, do not make reference in the report to "focus placeholders" just mention the "Focus box n".
+- in the Focus placeholder, make a mention to the prescribed sources used (no need to add the source before or after the placeholder)
+- do not make the placeholder on the exact same topic as the section or the sub-section where it is positioned, it has to be either:
     o a special case that deserves attention
     o a recent development / innovation
     o a theoretical drill-down
     o the guidance in terms of style and message to convey
     Note: Be specific if you want some particular point developped, keep it consistent across the report.
 - there should be between {round(pages/5,0)} and {round(pages/3.3,0)} of these focus placeholders within the report (all between introduction and conclusion)
+- 2 focus placeholders cannot be in the same section and should be a few pages apart in the report
 - Mention all the sources that should be used to generate this focus placeholder and list also the references that will be mentioned in the References section later (ex: [1], [2])
+Note: the Focus placeholders will then be processed separately by a llm to generate the specific code to display each of them so the instruction need to be clear enough.
 // Format:
 [[Focus Placeholder n: