pre release milestone: generate pdf

app.py

@@ -4,7 +4,7 @@ from config import STRONG_API_BASE, STRONG_API_KEY, STRONG_MODEL
 from util import is_valid_url
 from util import mylogger
 from util import stream_together
-from taskNonAI import extract_url, file_to_html
+from taskNonAI import extract_url, file_to_html, compile_pdf
 from taskAI import TaskAI
 ## load data
 from data_test import mock_jd, mock_cv
@@ -14,7 +14,7 @@ import gradio as gr
 from pypandoc.pandoc_download import download_pandoc
 ## std
 import os
-
+import json
 
 logger = mylogger(__name__,'%(asctime)s:%(levelname)s:%(message)s')
 info = logger.info
@@ -89,6 +89,11 @@ def finalize_letter_txt(api_base, api_key, api_model, debug_CoT, jd, cv):
     for result in gen:
         yield result
 
+def finalize_letter_pdf(meta_data, cover_letter_text):
+    pdf_context = json.loads(meta_data)
+    pdf_context["letter_body"] = cover_letter_text
+    return compile_pdf(pdf_context,tmpl_path="template_letter.tmpl",output_path=f"/tmp/cover_letter_{pdf_context['applicant_full_name']}_to_{pdf_context['company_full_name']}.pdf")
+
 with gr.Blocks(
     title=DEMO_TITLE,
     theme=gr.themes.Base(primary_hue="blue", secondary_hue="sky", neutral_hue="slate"),
@@ -175,11 +180,11 @@ with gr.Blocks(
     ).then(fn=run_compose, inputs=[strong_base, strong_key, strong_model, min_jd, min_cv], outputs=[debug_CoT]                      
     ).then(fn=lambda:gr.Accordion("Expert Zone", open=False),inputs=None, outputs=[expert_zone]
     ).then(fn=finalize_letter_txt, inputs=[cheap_base, cheap_key, cheap_model, debug_CoT, jd_info, cv_text], outputs=[cover_letter_text, debug_jobapp]
-    )
+    ).then(fn=finalize_letter_pdf, inputs=[debug_jobapp, cover_letter_text], outputs=[cover_letter_pdf])
 
 
 if __name__ == "__main__":
     init()
-    app.queue(max_size=10, default_concurrency_limit=1).launch(
+    app.queue(max_size=1, default_concurrency_limit=1).launch(
         show_error=True, debug=True, share=IS_SHARE
     )

data_test.py

@@ -65,3 +65,11 @@ References
 
 Available upon request.
 """
+pdf_context = {
+  "company_full_name": "Mastercard",
+  "job_title": "Project Management Intern",
+  "applicant_full_name": "Dorothy Gale",
+  "applicant_contact_information": "123 Main St., Emerald City, KS 12345, (123) 456-7890, dorothy@wizardofoz.com",
+  "letter_body": "text,\n\ntest test"
+}
+

requirements.txt

@@ -1,6 +1,7 @@
 # UI
 gradio>=4.0,<=5.0
 # external tool
+typst
 pypandoc
 shot-scraper
 # LLM related

taskNonAI.py

@@ -1,8 +1,14 @@
 import pypandoc
+import typst
 ## stdlib
 import subprocess
 import json
+import os
 from typing import Optional
+from string import Template
+
+from datetime import datetime
+
 
 def file_to_html(file_path: str) -> str:
     return pypandoc.convert_file(file_path, "html")
@@ -29,4 +35,24 @@ def extract_url(url: str) -> Optional[str]:
     except:
         raise Exception(
             f"Please try copy-paste as input. Failed to extract content from: {url}. Didn't find content from given URL!"
-        )
+        )
+
+def date():
+    current_date = datetime.now()
+    return current_date.strftime(
+        f"%B %d{'th' if 4 <= current_date.day <= 20 or 24 <= current_date.day <= 30 else ['st', 'nd', 'rd'][current_date.day % 10 - 1]} , %Y")
+
+def typst_escape(s):
+    return s.replace('@','\@').replace('#','\#')
+
+def compile_pdf(context: dict, tmpl_path: str, output_path="/tmp/cover_letter.pdf"):
+    with open(tmpl_path, "r", encoding='utf8') as f:
+        tmpl = Template(f.read())
+    context = {k: typst_escape(v) for k, v in context.items()}
+    context.update({'date_string': date()})
+    letter_typ = tmpl.safe_substitute(context)
+    with open('letter.typ', 'w', encoding='utf8') as f:
+        f.write(letter_typ)
+    typst.compile('letter.typ', output=output_path)
+    os.remove('letter.typ')
+    return output_path

template_base.typ

@@ -0,0 +1,62 @@
+// This function gets your whole document as its `body`
+// and formats it as a simple letter.
+#let letter(
+  // The letter's sender, which is display at the top of the page.
+  sender: none,
+
+  // The letter's recipient, which is displayed close to the top.
+  recipient: none,
+
+  // The date, displayed to the right.
+  date: none,
+
+  // The subject line.
+  subject: none,
+
+  // The name with which the letter closes.
+  name: none,
+
+  // The letter's content.
+  body
+) = {
+  // Configure page and text properties.
+  set page(paper: "us-letter", margin: (top: 2cm, bottom: 1.0cm))
+  set text(font: "PT Sans")
+
+  // Display sender at top of page. If there's no sender
+  // add some hidden text to keep the same spacing.
+  [== #name]
+  text(9pt, if sender == none {
+    hide("a")
+  } else {
+    sender
+  })
+
+  v(1.8cm)
+
+  // Display recipient.
+  recipient
+
+  v(0.5cm)
+
+  // Display date. If there's no date add some hidden
+  // text to keep the same spacing.
+  align(right, if date != none {
+    date
+  } else {
+    hide("a")
+  })
+
+  v(2cm)
+
+  // Add the subject line, if any.
+  if subject != none {
+    pad(right: 10%, strong(subject))
+  }
+
+  // Add body and name.
+  body
+  // v(1.25cm)
+  image("Signature.png", height: 7%, fit: "stretch")
+  name
+}

template_letter.tmpl

@@ -0,0 +1,15 @@
+#import "template_base.typ": *
+#show: letter.with(
+  sender: [
+     ${applicant_contact_information}
+  ],
+  recipient: [
+    Hiring Manager \
+    ${company_full_name} \
+  ],
+  date: [${date_string}],
+  subject: [Cover Letter for ${job_title}],
+  name: [${applicant_full_name}],
+)
+
+${letter_body}

test.py

@@ -1,5 +1,6 @@
 from taskAI import TaskAI
-from data_test import mock_jd, mock_cv
+from taskNonAI import compile_pdf
+from data_test import mock_jd, mock_cv, pdf_context
 from config_secret import api_test
 
 from llama_index.llms.openai_like import OpenAILike
@@ -27,6 +28,11 @@ def test_taskAI():
     for chunk in gen:
         print(chunk)
 
+def test_typst_pdf():
+    compile_pdf(tmpl_path='template_letter.tmpl',context=pdf_context)
+    os
+
 if __name__ == "__main__":
-    test_taskAI()
+    # test_taskAI()
     # integration()
+    test_typst_pdf()