Fix problems with prior commits

app.py

@@ -277,20 +277,32 @@ def generate_slide_deck(json_str: str):
             slides_template=pptx_template,
             output_file_path=path
         )
-
         _display_download_button(path)
-    except ValueError as ve:
-        st.error(
-            f"{APP_TEXT['json_parsing_error']}"
-            f"\n\nAdditional error info: {ve}"
-            f"\n\nHere are some sample instructions that you could try to possibly fix this error;"
-            f" if these don't work, try rephrasing or refreshing:"
-            f"\n\n"
-            "- Regenerate content and fix the JSON error."
-            "\n- Regenerate content and fix the JSON error. Quotes inside quotes should be escaped."
+    except ValueError:
+        # st.error(
+        #     f"{APP_TEXT['json_parsing_error']}"
+        #     f"\n\nAdditional error info: {ve}"
+        #     f"\n\nHere are some sample instructions that you could try to possibly fix this error;"
+        #     f" if these don't work, try rephrasing or refreshing:"
+        #     f"\n\n"
+        #     "- Regenerate content and fix the JSON error."
+        #     "\n- Regenerate content and fix the JSON error. Quotes inside quotes should be escaped."
+        # )
+        # logger.error('%s', APP_TEXT['json_parsing_error'])
+        # logger.error('Additional error info: %s', str(ve))
+        st.info(
+            'Encountered error while parsing JSON...will fix it and retry'
+        )
+        logger.debug(
+            'Caught ValueError: trying again after repairing JSON...'
+        )
+
+        pptx_helper.generate_powerpoint_presentation(
+            text_helper.fix_malformed_json(json_str),
+            slides_template=pptx_template,
+            output_file_path=path
         )
-        logger.error('%s', APP_TEXT['json_parsing_error'])
-        logger.error('Additional error info: %s', str(ve))
+        _display_download_button(path)
     except Exception as ex:
         st.error(APP_TEXT['content_generation_error'])
         logger.error('Caught a generic exception: %s', str(ex))

global_config.py

@@ -24,12 +24,12 @@ class GlobalConfig:
     APP_STRINGS_FILE = 'strings.json'
     PRELOAD_DATA_FILE = 'examples/example_02.json'
     SLIDES_TEMPLATE_FILE = 'langchain_templates/template_combined.txt'
-    JSON_TEMPLATE_FILE = 'langchain_templates/text_to_json_template_02.txt'
-    INITIAL_PROMPT_TEMPLATE = 'langchain_templates/chat_prompts/initial_template.txt'
-    REFINEMENT_PROMPT_TEMPLATE = 'langchain_templates/chat_prompts/refinement_template.txt'
+    # JSON_TEMPLATE_FILE = 'langchain_templates/text_to_json_template_02.txt'
+    INITIAL_PROMPT_TEMPLATE = 'langchain_templates/chat_prompts/initial_template_v2_steps.txt'
+    REFINEMENT_PROMPT_TEMPLATE = 'langchain_templates/chat_prompts/refinement_template_v2_steps.txt'
 
     PPTX_TEMPLATE_FILES = {
-        'Blank': {
+        'Basic': {
             'file': 'pptx_templates/Blank.pptx',
             'caption': 'A good start'
         },

helpers/pptx_helper.py

@@ -7,10 +7,21 @@ from typing import List, Tuple
 
 import json5
 import pptx
+from pptx.enum.shapes import MSO_AUTO_SHAPE_TYPE
 
 from global_config import GlobalConfig
 
 
+# English Metric Unit (used by PowerPoint) to inches
+EMU_TO_INCH_SCALING_FACTOR = 1.0 / 914400
+INCHES_1_5 = pptx.util.Inches(1.5)
+INCHES_1 = pptx.util.Inches(1)
+INCHES_0_5 = pptx.util.Inches(0.5)
+INCHES_0_4 = pptx.util.Inches(0.4)
+INCHES_0_3 = pptx.util.Inches(0.3)
+
+STEP_BY_STEP_PROCESS_MARKER = '>> '
+
 PATTERN = re.compile(r"^slide[ ]+\d+:", re.IGNORECASE)
 SAMPLE_JSON_FOR_PPTX = '''
 {
@@ -69,6 +80,10 @@ def generate_powerpoint_presentation(
     )
     presentation = pptx.Presentation(GlobalConfig.PPTX_TEMPLATE_FILES[slides_template]['file'])
 
+    slide_width_inch = EMU_TO_INCH_SCALING_FACTOR * presentation.slide_width
+    slide_height_inch = EMU_TO_INCH_SCALING_FACTOR * presentation.slide_height
+    logger.debug('Slide width: %f, height: %f', slide_width_inch, slide_height_inch)
+
     # The title slide
     title_slide_layout = presentation.slide_layouts[0]
     slide = presentation.slides.add_slide(title_slide_layout)
@@ -91,24 +106,21 @@ def generate_powerpoint_presentation(
     for a_slide in parsed_data['slides']:
         bullet_slide_layout = presentation.slide_layouts[1]
         slide = presentation.slides.add_slide(bullet_slide_layout)
-        shapes = slide.shapes
-
-        title_shape = shapes.title
-        body_shape = shapes.placeholders[1]
-        title_shape.text = remove_slide_number_from_heading(a_slide['heading'])
-        all_headers.append(title_shape.text)
-        text_frame = body_shape.text_frame
 
-        # The bullet_points may contain a nested hierarchy of JSON arrays
-        # In some scenarios, it may contain objects (dictionaries) because the LLM generated so
-        #  ^ The second scenario is not covered
-
-        flat_items_list = get_flat_list_of_contents(a_slide['bullet_points'], level=0)
-
-        for an_item in flat_items_list:
-            paragraph = text_frame.add_paragraph()
-            paragraph.text = an_item[0]
-            paragraph.level = an_item[1]
+        if not _handle_step_by_step_process(
+            slide=slide,
+            slide_json=a_slide,
+            slide_width_inch=slide_width_inch,
+            slide_height_inch=slide_height_inch,
+        ):
+            _handle_default_display(slide, a_slide)
+
+        _handle_key_message(
+            slide=slide,
+            slide_json=a_slide,
+            slide_width_inch=slide_width_inch,
+            slide_height_inch=slide_height_inch,
+        )
 
     # The thank-you slide
     last_slide_layout = presentation.slide_layouts[0]
@@ -142,28 +154,128 @@ def get_flat_list_of_contents(items: list, level: int) -> List[Tuple]:
     return flat_list
 
 
+def _handle_default_display(
+        slide: pptx.slide.Slide,
+        slide_json: dict,
+):
+    """
+    Display a list of text in a slide.
+
+    :param slide: The slide to be processed.
+    :param slide_json: The content of the slide as JSON data.
+    """
+
+    shapes = slide.shapes
+    title_shape = shapes.title
+    body_shape = shapes.placeholders[1]
+    title_shape.text = remove_slide_number_from_heading(slide_json['heading'])
+    text_frame = body_shape.text_frame
+
+    # The bullet_points may contain a nested hierarchy of JSON arrays
+    # In some scenarios, it may contain objects (dictionaries) because the LLM generated so
+    #  ^ The second scenario is not covered
+
+    flat_items_list = get_flat_list_of_contents(slide_json['bullet_points'], level=0)
+
+    for an_item in flat_items_list:
+        paragraph = text_frame.add_paragraph()
+        paragraph.text = an_item[0].removeprefix(STEP_BY_STEP_PROCESS_MARKER)
+        paragraph.level = an_item[1]
+
+
+def _handle_step_by_step_process(
+        slide: pptx.slide.Slide,
+        slide_json: dict,
+        slide_width_inch: float,
+        slide_height_inch: float
+) -> bool:
+    """
+    Add shapes to display a step-by-step process in the slide, if available.
+
+    :param slide: The slide to be processed.
+    :param slide_json: The content of the slide as JSON data.
+    :param slide_width_inch: The width of the slide in inches.
+    :param slide_height_inch: The height of the slide in inches.
+    :return True is this slide has a step-by-step process depiction; False otherwise.
+    """
+
+    if 'bullet_points' in slide_json and slide_json['bullet_points']:
+        steps = slide_json['bullet_points']
+
+        # Ensure that it is a single list of strings without any sub-list
+        for step in steps:
+            if not isinstance(step, str):
+                return False
+
+            if not step.startswith(STEP_BY_STEP_PROCESS_MARKER):
+                return False
+
+        shapes = slide.shapes
+        shapes.title.text = remove_slide_number_from_heading(slide_json['heading'])
+        n_steps = len(steps)
+
+        if 3 <= n_steps <= 4:
+            # Horizontal display
+            height = INCHES_1_5
+            width = pptx.util.Inches(slide_width_inch / n_steps - 0.01)
+            top = pptx.util.Inches(slide_height_inch / 2)
+            left = pptx.util.Inches((slide_width_inch - width.inches * n_steps) / 2 + 0.05)
+
+            for step in steps:
+                shape = shapes.add_shape(MSO_AUTO_SHAPE_TYPE.CHEVRON, left, top, width, height)
+                shape.text = step.removeprefix(STEP_BY_STEP_PROCESS_MARKER)
+                left += width - INCHES_0_4
+        elif n_steps > 4:
+            # Vertical display
+            height = pptx.util.Inches(0.65)
+            width = pptx.util.Inches(slide_width_inch * 2/ 3)
+            top = pptx.util.Inches(slide_height_inch / 4)
+            left = INCHES_1  # slide_width_inch - width.inches)
+
+            for step in steps:
+                shape = shapes.add_shape(MSO_AUTO_SHAPE_TYPE.PENTAGON, left, top, width, height)
+                shape.text = step.removeprefix(STEP_BY_STEP_PROCESS_MARKER)
+                top += height + INCHES_0_3
+                left += INCHES_0_5
+        else:
+            # Two steps -- probably not a process
+            return False
+
+    return True
+
+
+def _handle_key_message(
+        slide: pptx.slide.Slide,
+        slide_json: dict,
+        slide_width_inch: float,
+        slide_height_inch: float
+):
+    """
+    Add a shape to display the key message in the slide, if available.
+
+    :param slide: The slide to be processed.
+    :param slide_json: The content of the slide as JSON data.
+    :param slide_width_inch: The width of the slide in inches.
+    :param slide_height_inch: The height of the slide in inches.
+    """
+
+    if 'key_message' in slide_json and slide_json['key_message']:
+        height = pptx.util.Inches(1.6)
+        width = pptx.util.Inches(slide_width_inch / 2.3)
+        top = pptx.util.Inches(slide_height_inch - height.inches - 0.1)
+        left = pptx.util.Inches((slide_width_inch - width.inches) / 2)
+        shape = slide.shapes.add_shape(
+            MSO_AUTO_SHAPE_TYPE.ROUNDED_RECTANGLE,
+            left=left,
+            top=top,
+            width=width,
+            height=height
+        )
+        shape.text = slide_json['key_message']
+
+
 if __name__ == '__main__':
-    # bullets = [
-    #     'Description',
-    #     'Types',
-    #     [
-    #         'Type A',
-    #         'Type B'
-    #     ],
-    #     'Grand parent',
-    #     [
-    #         'Parent',
-    #         [
-    #             'Grand child'
-    #         ]
-    #     ]
-    # ]
-
-    # output = get_flat_list_of_contents(bullets, level=0)
-    # for x in output:
-    #     print(x)
-
-    json_data = '''
+    _JSON_DATA = '''
     {
     "title": "Understanding AI",
     "slides": [
@@ -174,7 +286,8 @@ if __name__ == '__main__':
                 [
                     "Importance of understanding AI"
                 ]
-            ]
+            ],
+            "key_message": ""
         },
         {
             "heading": "What is AI?",
@@ -188,7 +301,8 @@ if __name__ == '__main__':
                     ]
                 ],
                 "Differences between AI and machine learning"
-            ]
+            ],
+            "key_message": ""
         },
         {
             "heading": "How AI Works",
@@ -203,7 +317,19 @@ if __name__ == '__main__':
                     ]
                 ],
                 "How AI processes data"
-            ]
+            ],
+            "key_message": ""
+        },
+        {
+            "heading": "Building AI Models",
+            "bullet_points": [
+                ">> Collect data",
+                ">> Select model or architecture to use",
+                ">> Set appropriate parameters",
+                ">> Train model with data",
+                ">> Run inference",
+            ],
+            "key_message": ""
         },
         {
             "heading": "Pros of AI",
@@ -212,7 +338,8 @@ if __name__ == '__main__':
                 "Improved accuracy and precision",
                 "Enhanced decision-making capabilities",
                 "Personalized experiences"
-            ]
+            ],
+            "key_message": "AI can be used for many different purposes"
         },
         {
             "heading": "Cons of AI",
@@ -221,14 +348,16 @@ if __name__ == '__main__':
                 "Bias and discrimination",
                 "Privacy and security concerns",
                 "Dependence on technology"
-            ]
+            ],
+            "key_message": ""
         },
         {
             "heading": "Future Prospects of AI",
             "bullet_points": [
                 "Advancements in fields such as healthcare and finance",
                 "Increased use"
-            ]
+            ],
+            "key_message": ""
         }
     ]
 }'''
@@ -237,9 +366,10 @@ if __name__ == '__main__':
     path = pathlib.Path(temp.name)
 
     generate_powerpoint_presentation(
-        json5.loads(json_data),
+        json5.loads(_JSON_DATA),
         output_file_path=path,
-        slides_template='Blank'
+        slides_template='Basic'
     )
+    print(f'File path: {path}')
 
     temp.close()

helpers/text_helper.py

@@ -1,3 +1,6 @@
+import json_repair as jr
+
+
 def is_valid_prompt(prompt: str) -> bool:
     """
     Verify whether user input satisfies the concerned constraints.
@@ -46,11 +49,41 @@ def get_clean_json(json_str: str) -> str:
         # a new line or a closing bracket }
         prev_char = json_str[idx - 1]
 
-        if prev_char == '}':
-            response_cleaned = json_str[:idx]
-        elif prev_char == '\n' and json_str[idx - 2] == '}':
+        if (prev_char == '}') or (prev_char == '\n' and json_str[idx - 2] == '}'):
             response_cleaned = json_str[:idx]
 
         json_str = json_str[:idx]
 
     return response_cleaned
+
+
+def fix_malformed_json(json_str: str) -> str:
+    """
+    Try and fix the syntax error(s) in a JSON string.
+
+    :param json_str: The input JSON string.
+    :return: The fixed JSOn string.
+    """
+
+    return jr.repair_json(json_str, skip_json_loads=True)
+
+
+if __name__ == '__main__':
+    json1 = '''{
+    "key": "value"
+    }
+    '''
+    json2 = '''["Reason": "Regular updates help protect against known vulnerabilities."]'''
+    json3 = '''["Reason" Regular updates help protect against known vulnerabilities."]'''
+    json4 = '''
+    {"bullet_points": [
+        ">> Write without stopping or editing",
+        >> Set daily writing goals and stick to them,
+        ">> Allow yourself to make mistakes"
+    ],}
+    '''
+
+    print(fix_malformed_json(json1))
+    print(fix_malformed_json(json2))
+    print(fix_malformed_json(json3))
+    print(fix_malformed_json(json4))

langchain_templates/chat_prompts/initial_template_v2_steps.txt

@@ -0,0 +1,59 @@
+You are a helpful, intelligent chatbot. Create the slides for a presentation on the given topic.
+Include main headings for each slide, detailed bullet points for each slide.
+Add relevant content to each slide.
+The content of each slide should be verbose, descriptive, and very detailed.
+If relevant, add one or two examples to illustrate the concept.
+Unless explicitly specified with the topic, create about 10 slides.
+
+
+### Topic:
+{question}
+
+
+The output must be only a valid and syntactically correct JSON adhering to the following schema:
+{{
+    "title": "Presentation Title",
+    "slides": [
+        {{
+            "heading": "Heading for the First Slide",
+            "bullet_points": [
+                "First bullet point",
+                [
+                    "Sub-bullet point 1",
+                    "Sub-bullet point 2"
+                ],
+                "Second bullet point"
+            ],
+            "key_message": ""
+        }},
+        {{
+            "heading": "Heading for the Second Slide",
+            "bullet_points": [
+                "First bullet point",
+                "Second bullet item",
+                "Third bullet point"
+            ],
+            "key_message": "The key message conveyed in this slide"
+        }},
+        {{
+            "heading": "A slide that describes a step-by-step/sequential process",
+            "bullet_points": [
+                ">> The first step of the process (begins with special marker >>)",
+                ">> A second step (begins with >>)",
+                ">> Third step",
+            ],
+            "key_message": ""
+        }}
+    ]
+}}
+
+
+### Some more hints on the slide content and JSON output format:
+- For two or three important slides, generate the key message that those slides convey and assign
+them to the `key_message` elements of JSON output.
+- Identify if a slide describes a step-by-step/sequential process, then begin the bullet points
+with a special marker >>. Limit this to max two or three slides.
+
+
+### Output:
+```json

langchain_templates/chat_prompts/refinement_template_v2_steps.txt

@@ -0,0 +1,70 @@
+You are a helpful, intelligent chatbot. You follow instructions to refine an existing slide deck.
+A list of user instructions is provided below in sequential order -- from the oldest to the latest.
+The previously generated content of the slide deck in JSON format is also provided.
+Follow the instructions to revise the content of the previously generated slides of the presentation on the given topic.
+Include main headings for each slide, detailed bullet points for each slide.
+Add relevant content to each slide.
+The content of the slides should be descriptive, verbose, and detailed.
+If relevant, add one or two examples to illustrate the concept.
+Unless explicitly specified with the topic, create about 10 slides.
+You also fix any syntax error that may be present in the JSON-formatted content.
+
+A slide that describes a step-by-step/sequential process begins the bullet points
+with a special marker >>
+
+
+### List of instructions:
+{instructions}
+
+
+### Previously generated slide deck content as JSON:
+{previous_content}
+
+
+The output must be only a valid and syntactically correct JSON adhering to the following schema:
+{{
+    "title": "Presentation Title",
+    "slides": [
+        {{
+            "heading": "Heading for the First Slide",
+            "bullet_points": [
+                "First bullet point",
+                [
+                    "Sub-bullet point 1",
+                    "Sub-bullet point 2"
+                ],
+                "Second bullet point"
+            ],
+            "key_message": ""
+        }},
+        {{
+            "heading": "Heading for the Second Slide",
+            "bullet_points": [
+                "First bullet point",
+                "Second bullet item",
+                "Third bullet point"
+            ],
+            "key_message": "The key message conveyed in this slide"
+        }},
+        {{
+            "heading": "A slide that describes a step-by-step/sequential process",
+            "bullet_points": [
+                ">> The first step of the process (begins with special marker >>)",
+                ">> A second step (begins with >>)",
+                ">> Third step",
+            ],
+            "key_message": ""
+        }}
+    ]
+}}
+
+
+### Some more hints on the slide content and JSON output format:
+- For two or three important slides, generate the key message that those slides convey and assign
+them to the `key_message` elements of JSON output.
+- Identify if a slide describes a step-by-step/sequential process, then begin the bullet points
+with a special marker >>. Limit this to max two or three slides.
+
+
+### Output:
+```json

requirements.txt

@@ -1,6 +1,14 @@
+aiohttp==3.9.5
 python-dotenv[cli]~=1.0.0
-langchain~=0.1.13
-# huggingface_hub
+gitpython==3.1.43
+json_repair==0.15.3
+idna==3.7
+jinja2==3.1.3
+Pillow==10.3.0
+pyarrow~=16.0.0
+pydantic==2.4.0
+langchain~=0.1.16
+langchain-core~=0.1.46
 streamlit~=1.32.2
 
 python-pptx