Upload folder using huggingface_hub

app.py

@@ -1,42 +1,70 @@
 import gradio as gr
+from pypdf import PdfReader
 
+from explanations import explanation1, explanation2, explanation3
 from transcript_transformer import TranscriptTransformer
 
 transformer = TranscriptTransformer()
 
+def process_pdf(pdf_file):
+    pdf_reader = PdfReader(pdf_file.name)
+    text = ""
+    for page in pdf_reader.pages:
+        text += page.extract_text()
+    return text
 
-def process_transcript(transcript: str, duration: int):
+def process_transcript(transcript_text: str, pdf_file, duration: int):
     yield gr.update(value="Building the lecture..", visible=True)
-
+    
+    # Use PDF content if provided, otherwise use transcript text
+    if pdf_file:
+        transcript = process_pdf(pdf_file)
+    else:
+        transcript = transcript_text
+        
     transformed_transcript = transformer.generate_lecture(transcript, duration)
-
     yield gr.update(value=transformed_transcript, visible=True)
 
 
-demo = gr.Interface(
-    fn=process_transcript,
-    inputs=[
-        gr.Textbox(
-            label="Input Transcript",
-            placeholder="Paste your transcript here...",
-            lines=10,
-        ),
-        gr.Slider(
-            minimum=15,
-            maximum=60,
-            value=30,
-            step=15,
-            label="Lecture Duration (minutes)",
-        ),
-    ],
-    outputs=gr.Markdown(label="Transformed Teaching Transcript"),
-    title="Transcript to Teaching Material Transformer",
-    description="""Transform transcripts into teaching materials.
-    The output will be formatted as a complete lecture with clear sections,
-    examples, and interactive elements.""",
-    theme="default",
-)
-
-
-if __name__ == "__main__":
-    demo.launch()
+with gr.Blocks() as demo:
+    accordion1 = gr.Accordion("How prompts were engineered and refined?", open=False)
+    with accordion1:
+        gr.Markdown(explanation1)
+
+    accordion2 = gr.Accordion("Challenges faced", open=False)
+    with accordion2:
+        gr.Markdown(explanation2)
+
+    accordion3 = gr.Accordion("How the system can be extended or scaled?", open=False)
+    with accordion3:
+        gr.Markdown(explanation3)
+
+    gr.Interface(
+        fn=process_transcript,
+        inputs=[
+            gr.Textbox(
+                label="Input Transcript",
+                placeholder="Paste your transcript here...",
+                lines=10,
+            ),
+            gr.File(
+                label="Or Upload PDF",
+                file_types=[".pdf"],
+            ),
+            gr.Slider(
+                minimum=15,
+                maximum=60,
+                value=30,
+                step=15,
+                label="Lecture Duration (minutes)",
+            ),
+        ],
+        outputs=gr.Markdown(label="Transformed Teaching Transcript"),
+        title="Transcript to Teaching Material Transformer",
+        description="""Transform transcripts into teaching materials.
+        The output will be formatted as a complete lecture with clear sections,
+        examples, and interactive elements.""",
+        theme="default",
+    )
+
+demo.launch()

explanations.py

@@ -0,0 +1,46 @@
+explanation1 = """
+I used two system prompts: one for summarizing the chunks if they surpass the amount of tokens
+of gpt-4o-mini (the LLM model used for this app) and another one for the final response.
+These were just to guide the AI that it is an expert educational content creator and to act like one.
+
+I also used two user prompts for the same reason. One for summarizing chunks (if needed) and another
+one for the final response. The user prompt for summarizing chunks includes the context of the previous
+chunk and the current chunk. The user prompt for the final response includes the duration of the lecture
+and the word count of the transcript. The word count of the transcript assumes that an average paced lecturer
+speaks 130 words per minute.
+
+They system prompt to generate the lecture tells the AI that it must focus on clarity,
+logical flow, and educational value.
+
+The user prompt to generate the lecture clearly describes the requirements to give structure to
+the lecture and to make it more understandable, stating that it must:
+
+1. Create a clear introduction that sets context and learning objectives
+2. Organize the content into logical sections with clear headings
+3. Include practical examples and real-world applications
+4. Add discussion questions or interactive elements
+5. Conclude with a summary and key takeaways
+6. Target approximately {word_count} words
+
+and to format the output in markdown with clear section headers and proper spacing.
+"""
+
+explanation2 = """
+The main challenge was to maintain the context of the different
+chunks of the transcript and then combine them (if the total amount of tokens of the transcript surpasses the limit). 
+I had to make sure that the AI understands the context of the previous chunk and then summarize the current chunk
+accordingly.
+"""
+
+explanation3 = """
+The system can be extended by adding dynamic model selection.
+This means using other LLMs like Claude or open source LLMs from HuggingFace to see different results and
+stay with the one that is the best in terms of costs and performance.
+
+Another idea is to generate content according to the level of difficulty.
+For example, content for beginners, intermediate learners, or advanced audiences.
+
+Also the system could handle videos, written articles, or interactive presentations.
+
+Finally, the system could be deployed as an API to integrate it with other services.
+"""

transcript_transformer.py

@@ -5,14 +5,19 @@ load_dotenv()
 
 
 class TranscriptTransformer:
-    def __init__(self):
-        # System prompt template for consistent output
-        self.system_prompt = """You are an expert educational content creator. 
+    def __init__(self, model="gpt-4o-mini", max_tokens=128000):
+        self.model = model
+        self.max_model_tokens = max_tokens
+        self.final_response_system_prompt = """You are an expert educational content creator. 
         Your task is to transform informal transcripts into structured, engaging teaching materials.
         Focus on clarity, logical flow, and educational value."""
 
-        # Template for the lecture structure
-        self.lecture_template = """Transform the following transcript into a structured {duration}-minute lecture.
+        self.summarizer_system_prompt = """You are an expert educational content creator.
+        Your task is to summarize the following transcript chunk.
+        Ensure that the summary is concise and captures the main points.
+        """
+
+        self.final_response_user_prompt = """Transform the following transcript into a structured {duration}-minute lecture.
         
         Requirements:
         1. Create a clear introduction that sets context and learning objectives
@@ -25,7 +30,11 @@ class TranscriptTransformer:
         Format the output in markdown with clear section headers and proper spacing.
         """
 
-        self.max_model_tokens = 128000
+        self.summarizer_user_prompt = """
+        {context}
+        Summarize the following transcript chunk:
+        {chunk}
+        """
 
     def split_text_into_chunks(self, text: str) -> list[str]:
         """Split the text into chunks that fit within the token limit."""
@@ -55,17 +64,11 @@ class TranscriptTransformer:
             else ""
         )
 
-        prompt = f"""
-        {context}
-        Summarize the following transcript chunk:
-        {chunk}
-        """
-
         response = openai.chat.completions.create(
-            model="gpt-4o-mini",
+            model=self.model,
             messages=[
-                {"role": "system", "content": "You are a highly skilled educator AI."},
-                {"role": "user", "content": prompt},
+                {"role": "system", "content": self.summarizer_system_prompt},
+                {"role": "user", "content": self.summarizer_system_prompt.format(context=context, chunk=chunk)},
             ],
             max_tokens=max_output_tokens,
             temperature=0.7,
@@ -96,11 +99,11 @@ class TranscriptTransformer:
         summarized_transcript = self.summarize_text(
             raw_text, max_output_tokens=max_output_tokens
         )
-        full_text = f"{self.lecture_template.format(duration=lecture_duration, word_count=max_output_tokens)}\n\nTranscript:\n{summarized_transcript}"
+        full_text = f"{self.final_response_user_prompt.format(duration=lecture_duration, word_count=max_output_tokens)}\n\nTranscript:\n{summarized_transcript}"
         final_response = openai.chat.completions.create(
             model="gpt-4o-mini",
             messages=[
-                {"role": "system", "content": self.system_prompt},
+                {"role": "system", "content": self.final_response_system_prompt},
                 {"role": "user", "content": full_text},
             ],
             max_tokens=max_output_tokens,