Spaces:
Running
Running
SyedAzlanzar
commited on
Commit
·
67ba1ec
1
Parent(s):
c232ec4
@feat : cover letter enhanced, cl upload file, comment out some hf code.
Browse files- README.md +5 -0
- app/api/routes.py +42 -17
- app/main.py +0 -1
- app/services/generator.py +318 -195
- app/services/hf_storage_service.py +2 -2
- app/services/pdf_creator.py +10 -3
- requirements.txt +4 -1
README.md
CHANGED
|
@@ -8,3 +8,8 @@ pinned: false
|
|
| 8 |
---
|
| 9 |
|
| 10 |
Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 8 |
---
|
| 9 |
|
| 10 |
Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
|
| 11 |
+
|
| 12 |
+
|
| 13 |
+
<!-- .\venv\Scripts\activate.bat -->
|
| 14 |
+
<!-- pip install -r requirements.txt -->
|
| 15 |
+
<!-- uvicorn app.main:app --reload -->
|
app/api/routes.py
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
from fastapi import APIRouter, HTTPException
|
| 2 |
from app.models.schema import GenerateRequest, GenerateResponse
|
| 3 |
-
from app.services.generator import
|
| 4 |
from app.services.pdf_creator import save_pdf
|
| 5 |
from app.services.resume_parser import extract_resume_text
|
| 6 |
from app.utils.file_utils import generate_unique_filename
|
|
@@ -19,27 +19,27 @@ async def generate_cover_letter_api(data: GenerateRequest):
|
|
| 19 |
raise HTTPException(status_code=400, detail="Job detail is too long")
|
| 20 |
|
| 21 |
resume_text = extract_resume_text(data.resume_path)
|
| 22 |
-
letter_text = await
|
| 23 |
-
|
| 24 |
-
md_cover_letter = build_cover_letter_md(
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
|
| 33 |
-
)
|
| 34 |
|
| 35 |
-
text_cover_letter = convert_md_to_text(md_cover_letter)
|
| 36 |
|
| 37 |
filename = generate_unique_filename()
|
| 38 |
-
pdf_path = save_pdf(text_cover_letter, filename)
|
| 39 |
|
| 40 |
return GenerateResponse(
|
| 41 |
letter=letter_text,
|
| 42 |
-
pdf_url=
|
| 43 |
)
|
| 44 |
|
| 45 |
except HTTPException as http_exc:
|
|
@@ -72,4 +72,29 @@ async def upload_resume(resume: UploadFile = File(...)):
|
|
| 72 |
"success": False,
|
| 73 |
"error": str(e)
|
| 74 |
}
|
| 75 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
from fastapi import APIRouter, HTTPException
|
| 2 |
from app.models.schema import GenerateRequest, GenerateResponse
|
| 3 |
+
from app.services.generator import coverLetterGenerativeAIBot
|
| 4 |
from app.services.pdf_creator import save_pdf
|
| 5 |
from app.services.resume_parser import extract_resume_text
|
| 6 |
from app.utils.file_utils import generate_unique_filename
|
|
|
|
| 19 |
raise HTTPException(status_code=400, detail="Job detail is too long")
|
| 20 |
|
| 21 |
resume_text = extract_resume_text(data.resume_path)
|
| 22 |
+
letter_text = await coverLetterGenerativeAIBot(data, resume_text)
|
| 23 |
+
|
| 24 |
+
# md_cover_letter = build_cover_letter_md(
|
| 25 |
+
# your_name=data.full_name,
|
| 26 |
+
# postal_code=data.postal_code,
|
| 27 |
+
# city=data.city,
|
| 28 |
+
# email=data.email,
|
| 29 |
+
# phone=data.phone_number,
|
| 30 |
+
# job_title=data.job_title,
|
| 31 |
+
# company_name=data.company_name,
|
| 32 |
+
# generated_paragraphs=letter_text
|
| 33 |
+
# )
|
| 34 |
|
| 35 |
+
# text_cover_letter = convert_md_to_text(md_cover_letter)
|
| 36 |
|
| 37 |
filename = generate_unique_filename()
|
| 38 |
+
# pdf_path = save_pdf(text_cover_letter, filename)
|
| 39 |
|
| 40 |
return GenerateResponse(
|
| 41 |
letter=letter_text,
|
| 42 |
+
pdf_url=""
|
| 43 |
)
|
| 44 |
|
| 45 |
except HTTPException as http_exc:
|
|
|
|
| 72 |
"success": False,
|
| 73 |
"error": str(e)
|
| 74 |
}
|
| 75 |
+
|
| 76 |
+
@router.post("/upload-file")
|
| 77 |
+
async def upload_file(pdf: UploadFile = File(...)):
|
| 78 |
+
try:
|
| 79 |
+
# Read resume content
|
| 80 |
+
pdf_content = await pdf.read()
|
| 81 |
+
|
| 82 |
+
# Upload to HuggingFace Hub
|
| 83 |
+
pdf_url = storage_service.upload_file_to_hf(
|
| 84 |
+
file_content=pdf_content,
|
| 85 |
+
folder="cover-letters",
|
| 86 |
+
filename=pdf.filename
|
| 87 |
+
)
|
| 88 |
+
|
| 89 |
+
return {
|
| 90 |
+
"success": True,
|
| 91 |
+
"url": pdf_url
|
| 92 |
+
}
|
| 93 |
+
|
| 94 |
+
except Exception as e:
|
| 95 |
+
return {
|
| 96 |
+
"success": False,
|
| 97 |
+
"error": str(e)
|
| 98 |
+
}
|
| 99 |
+
|
| 100 |
+
|
app/main.py
CHANGED
|
@@ -38,7 +38,6 @@ async def log_requests(request: Request, call_next):
|
|
| 38 |
|
| 39 |
@app.get("/")
|
| 40 |
def ping():
|
| 41 |
-
print()
|
| 42 |
return {"status": "ok"}
|
| 43 |
|
| 44 |
|
|
|
|
| 38 |
|
| 39 |
@app.get("/")
|
| 40 |
def ping():
|
|
|
|
| 41 |
return {"status": "ok"}
|
| 42 |
|
| 43 |
|
app/services/generator.py
CHANGED
|
@@ -2,238 +2,361 @@ from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
|
|
| 2 |
import torch
|
| 3 |
from app.models.schema import GenerateRequest
|
| 4 |
import os
|
|
|
|
|
|
|
| 5 |
|
| 6 |
|
| 7 |
-
os.environ["TRANSFORMERS_CACHE"] = "/code/backend/transformers_cache"
|
| 8 |
-
os.makedirs(os.environ["TRANSFORMERS_CACHE"], exist_ok=True)
|
| 9 |
|
| 10 |
|
| 11 |
-
# Initialization model immediately after server is Up!
|
| 12 |
-
device = "cuda" if torch.cuda.is_available() else "cpu"
|
| 13 |
-
dtype = torch.float16 if device == "cuda" else torch.float32
|
| 14 |
|
| 15 |
-
# TinyLlama/TinyLlama-1.1B-Chat-v1.0
|
| 16 |
-
model_name = "Qwen/Qwen2.5-0.5B-Instruct"
|
| 17 |
-
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
| 18 |
-
model = AutoModelForCausalLM.from_pretrained(
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
)
|
| 23 |
|
| 24 |
-
model = model.to(device)
|
| 25 |
|
| 26 |
|
| 27 |
-
async def generate_cover_letter(data: GenerateRequest, resume_text: str) -> str:
|
| 28 |
-
|
| 29 |
|
| 30 |
-
|
| 31 |
|
| 32 |
|
| 33 |
-
|
| 34 |
-
|
| 35 |
|
| 36 |
-
|
| 37 |
|
| 38 |
-
|
| 39 |
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
| 43 |
|
| 44 |
|
| 45 |
-
|
| 46 |
-
|
| 47 |
|
| 48 |
-
|
| 49 |
-
|
| 50 |
|
| 51 |
-
|
| 52 |
-
|
| 53 |
-
|
| 54 |
-
## Output Format
|
| 55 |
-
|
| 56 |
-
Provide only the paragraph content in this format:
|
| 57 |
-
|
| 58 |
-
Dear {data.company_name},
|
| 59 |
-
|
| 60 |
-
## Core Writing Instructions
|
| 61 |
-
|
| 62 |
-
### Content Strategy (German Market Focus)
|
| 63 |
-
|
| 64 |
-
1. **Opening:** State the position you're applying for and briefly explain your interest in the company mentioned in the job details
|
| 65 |
-
2. **Body:** Focus on what value you bring to the company (not what you can gain from them)
|
| 66 |
-
3. **Skills Gap Handling:** If you lack certain required skills, show eagerness to learn and provide examples of quick skill acquisition
|
| 67 |
-
4. **Experience Mismatch:** If years of experience don't perfectly match (e.g., job asks 5+, you have ~4.5), emphasize the quality and relevance of your experience
|
| 68 |
-
5. **Closing:** Professional closing with availability and next steps
|
| 69 |
-
|
| 70 |
-
### Language Requirements
|
| 71 |
-
|
| 72 |
-
- **Tone:** Professional and formal (German market prefers formal tone over conversational)
|
| 73 |
-
- **Length:** Concise and focused (3 paragraphs preferred for German market, maximum 4 paragraphs)
|
| 74 |
-
- **Total word count:** 250-400 words maximum
|
| 75 |
-
- **Per paragraph:** 70-100 words (3-4 sentences each)
|
| 76 |
-
- **Opening:** 50-70 words
|
| 77 |
-
- **Body:** 80-100 words
|
| 78 |
-
- **Closing:** 50-70 words
|
| 79 |
-
- **Format:** Use continuous text only - no bullet points or lists within the paragraph body
|
| 80 |
-
- **Avoid These Overused Words:**
|
| 81 |
-
- Robust → use "strong," "reliable," or "sturdy"
|
| 82 |
-
- Keen → use "eager," "interested," or "enthusiastic"
|
| 83 |
-
- Scalable → use "that can handle growth"
|
| 84 |
-
- Leverage → use "use"
|
| 85 |
-
- Utilize → use "use"
|
| 86 |
-
- Innovative → describe specific innovation
|
| 87 |
-
- Optimal → use "best," "ideal," or "most effective"
|
| 88 |
-
- Capability → use "ability," "feature," or "function"
|
| 89 |
-
- Interface → use "connect," "work with," or describe interaction
|
| 90 |
-
- Efficient → use "quick," "streamlined," or describe time savings
|
| 91 |
-
- Paradigm → use "model," "approach," or "way of thinking"
|
| 92 |
-
- Facilitate → use "help," "make easier," or "support"
|
| 93 |
-
- Optimize → use "improve," "refine," or describe enhancement
|
| 94 |
-
|
| 95 |
-
### Specific Content Elements to Include When Relevant
|
| 96 |
|
| 97 |
-
|
| 98 |
|
| 99 |
-
|
| 100 |
-
- Address language barrier concerns proactively
|
| 101 |
|
| 102 |
-
|
| 103 |
|
| 104 |
-
|
| 105 |
-
- Highlight strong fundamentals that enable quick learning
|
| 106 |
-
- Provide examples of past rapid skill acquisition
|
| 107 |
|
| 108 |
-
|
| 109 |
-
|
| 110 |
-
|
| 111 |
-
|
| 112 |
-
|
| 113 |
-
|
| 114 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 115 |
|
| 116 |
-
|
| 117 |
-
|
| 118 |
-
3. **Authenticity:** Write like a real human - natural flow, simple language, avoid overly polished corporate speak
|
| 119 |
-
4. **Specificity:** Reference specific company details, role requirements, and technologies mentioned in the JD
|
| 120 |
|
| 121 |
-
|
|
|
|
|
|
|
|
|
|
| 122 |
|
| 123 |
-
|
|
|
|
| 124 |
|
| 125 |
-
|
| 126 |
-
|
| 127 |
-
|
| 128 |
-
|
| 129 |
-
|
| 130 |
-
|
| 131 |
-
|
| 132 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 133 |
|
| 134 |
-
|
|
|
|
| 135 |
|
| 136 |
-
|
| 137 |
|
|
|
|
|
|
|
|
|
|
| 138 |
|
| 139 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 140 |
|
|
|
|
|
|
|
| 141 |
messages = [
|
| 142 |
-
|
|
|
|
| 143 |
]
|
| 144 |
-
# Apply chat template
|
| 145 |
-
text = tokenizer.apply_chat_template(
|
| 146 |
-
messages,
|
| 147 |
-
tokenize=False,
|
| 148 |
-
add_generation_prompt=True
|
| 149 |
-
)
|
| 150 |
-
|
| 151 |
-
# Tokenize input
|
| 152 |
-
model_inputs = tokenizer(
|
| 153 |
-
[text],
|
| 154 |
-
return_tensors="pt",
|
| 155 |
-
padding=True,
|
| 156 |
-
truncation=True,
|
| 157 |
-
max_length=16000
|
| 158 |
-
).to(model.device)
|
| 159 |
-
|
| 160 |
-
# Generate with parameters optimized for markdown
|
| 161 |
-
generated_ids = model.generate(
|
| 162 |
-
**model_inputs,
|
| 163 |
-
max_new_tokens=7192, # Increased for longer markdown content
|
| 164 |
-
do_sample=True,
|
| 165 |
-
temperature=0.8, # Slightly lower for more consistent formatting
|
| 166 |
-
top_p=0.9,
|
| 167 |
-
top_k=50,
|
| 168 |
-
repetition_penalty=1.15, # Higher to avoid repetitive formatting
|
| 169 |
-
eos_token_id=tokenizer.eos_token_id,
|
| 170 |
-
pad_token_id=tokenizer.pad_token_id
|
| 171 |
-
)
|
| 172 |
-
|
| 173 |
-
generated_only_ids = generated_ids[:, model_inputs.input_ids.shape[1]:]
|
| 174 |
-
|
| 175 |
|
| 176 |
-
#
|
| 177 |
-
|
| 178 |
-
|
| 179 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 180 |
)
|
| 181 |
|
| 182 |
-
|
| 183 |
-
|
| 184 |
-
|
| 185 |
-
|
| 186 |
-
except Exception as e:
|
| 187 |
-
return {"error": str(e)}
|
| 188 |
-
|
| 189 |
-
|
| 190 |
-
async def job_details_alteration(job_details:str) -> str:
|
| 191 |
-
try:
|
| 192 |
-
prompt = f"""
|
| 193 |
-
job_description: {job_details}
|
| 194 |
-
|
| 195 |
-
Your task: Extract only the important parts and rewrite them clearly into 4 sections:
|
| 196 |
-
- **Responsibilities** (what the candidate will do)
|
| 197 |
-
- **Requirements / Qualifications** (skills, education, experience needed)
|
| 198 |
-
- **Who You Are** (traits, mindset, culture fit)
|
| 199 |
-
- **Preferred Candidate** (optional nice-to-have skills or experience)
|
| 200 |
|
| 201 |
-
Rules:
|
| 202 |
-
- Remove irrelevant parts such as "About Us", "Why Join Us", "Perks/Benefits", or generic company marketing.
|
| 203 |
-
- Keep the output concise and professional.
|
| 204 |
-
- If a section is not found, skip it. """
|
| 205 |
|
| 206 |
-
messages = [
|
| 207 |
-
{"role": "system", "content": prompt}
|
| 208 |
-
]
|
| 209 |
-
text = tokenizer.apply_chat_template(
|
| 210 |
-
messages,
|
| 211 |
-
tokenize=False,
|
| 212 |
-
add_generation_prompt=True
|
| 213 |
-
)
|
| 214 |
-
model_inputs = tokenizer(
|
| 215 |
-
[text],
|
| 216 |
-
return_tensors="pt",
|
| 217 |
-
padding=True,
|
| 218 |
-
truncation=True,
|
| 219 |
-
max_length=2048
|
| 220 |
-
).to(model.device)
|
| 221 |
-
generated_ids = model.generate(
|
| 222 |
-
**model_inputs,
|
| 223 |
-
max_new_tokens=8192,
|
| 224 |
-
do_sample=True,
|
| 225 |
-
temperature=0.5,
|
| 226 |
-
top_p=0.9,
|
| 227 |
-
top_k=50,
|
| 228 |
-
repetition_penalty=1.15,
|
| 229 |
-
eos_token_id=tokenizer.eos_token_id,
|
| 230 |
-
pad_token_id=tokenizer.pad_token_id
|
| 231 |
-
)
|
| 232 |
-
generated_only_ids = generated_ids[:, model_inputs.input_ids.shape[1]:]
|
| 233 |
-
response = tokenizer.decode(
|
| 234 |
-
generated_only_ids[0],
|
| 235 |
-
skip_special_tokens=True
|
| 236 |
-
)
|
| 237 |
-
return response.strip()
|
| 238 |
except Exception as e:
|
| 239 |
-
|
|
|
|
|
|
| 2 |
import torch
|
| 3 |
from app.models.schema import GenerateRequest
|
| 4 |
import os
|
| 5 |
+
from openai import OpenAI
|
| 6 |
+
import tiktoken
|
| 7 |
|
| 8 |
|
| 9 |
+
# os.environ["TRANSFORMERS_CACHE"] = "/code/backend/transformers_cache"
|
| 10 |
+
# os.makedirs(os.environ["TRANSFORMERS_CACHE"], exist_ok=True)
|
| 11 |
|
| 12 |
|
| 13 |
+
# # Initialization model immediately after server is Up!
|
| 14 |
+
# device = "cuda" if torch.cuda.is_available() else "cpu"
|
| 15 |
+
# dtype = torch.float16 if device == "cuda" else torch.float32
|
| 16 |
|
| 17 |
+
# # TinyLlama/TinyLlama-1.1B-Chat-v1.0
|
| 18 |
+
# model_name = "Qwen/Qwen2.5-0.5B-Instruct"
|
| 19 |
+
# tokenizer = AutoTokenizer.from_pretrained(model_name)
|
| 20 |
+
# model = AutoModelForCausalLM.from_pretrained(
|
| 21 |
+
# model_name,
|
| 22 |
+
# dtype="auto",
|
| 23 |
+
# device_map="auto"
|
| 24 |
+
# )
|
| 25 |
|
| 26 |
+
# model = model.to(device)
|
| 27 |
|
| 28 |
|
| 29 |
+
# async def generate_cover_letter(data: GenerateRequest, resume_text: str) -> str:
|
| 30 |
+
# try:
|
| 31 |
|
| 32 |
+
# altered_job_details = await job_details_alteration(data.job_details)
|
| 33 |
|
| 34 |
|
| 35 |
+
# prompt = f"""
|
| 36 |
+
# # AI Writer Bot Prompt
|
| 37 |
|
| 38 |
+
# You are an expert writer that creates personalized, human-written paragraphs based on job descriptions and CV information. Your goal is to craft authentic paragraphs that sound like they're written by a Pakistani professional with 6.5 IELTS band score - natural, clear, but not overly sophisticated or AI-generated.
|
| 39 |
|
| 40 |
+
# ## Input Requirements
|
| 41 |
|
| 42 |
+
# - ** Job Title :** Exact title of the position being applied for
|
| 43 |
+
# - **Job Description (JD):** Complete job posting with requirements, responsibilities, and company information
|
| 44 |
+
# - **CV/Resume:** Tailored resume specifically prepared for the target role
|
| 45 |
|
| 46 |
|
| 47 |
+
# Job Title:
|
| 48 |
+
# {data.job_title}
|
| 49 |
|
| 50 |
+
# Job Description:
|
| 51 |
+
# {altered_job_details}
|
| 52 |
|
| 53 |
+
# Resume Content:
|
| 54 |
+
# {resume_text}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 55 |
|
| 56 |
+
# ## Output Format
|
| 57 |
|
| 58 |
+
# Provide only the paragraph content in this format:
|
|
|
|
| 59 |
|
| 60 |
+
# Dear {data.company_name},
|
| 61 |
|
| 62 |
+
# ## Core Writing Instructions
|
|
|
|
|
|
|
| 63 |
|
| 64 |
+
# ### Content Strategy (German Market Focus)
|
| 65 |
+
|
| 66 |
+
# 1. **Opening:** State the position you're applying for and briefly explain your interest in the company mentioned in the job details
|
| 67 |
+
# 2. **Body:** Focus on what value you bring to the company (not what you can gain from them)
|
| 68 |
+
# 3. **Skills Gap Handling:** If you lack certain required skills, show eagerness to learn and provide examples of quick skill acquisition
|
| 69 |
+
# 4. **Experience Mismatch:** If years of experience don't perfectly match (e.g., job asks 5+, you have ~4.5), emphasize the quality and relevance of your experience
|
| 70 |
+
# 5. **Closing:** Professional closing with availability and next steps
|
| 71 |
+
|
| 72 |
+
# ### Language Requirements
|
| 73 |
+
|
| 74 |
+
# - **Tone:** Professional and formal (German market prefers formal tone over conversational)
|
| 75 |
+
# - **Length:** Concise and focused (3 paragraphs preferred for German market, maximum 4 paragraphs)
|
| 76 |
+
# - **Total word count:** 250-400 words maximum
|
| 77 |
+
# - **Per paragraph:** 70-100 words (3-4 sentences each)
|
| 78 |
+
# - **Opening:** 50-70 words
|
| 79 |
+
# - **Body:** 80-100 words
|
| 80 |
+
# - **Closing:** 50-70 words
|
| 81 |
+
# - **Format:** Use continuous text only - no bullet points or lists within the paragraph body
|
| 82 |
+
# - **Avoid These Overused Words:**
|
| 83 |
+
# - Robust → use "strong," "reliable," or "sturdy"
|
| 84 |
+
# - Keen → use "eager," "interested," or "enthusiastic"
|
| 85 |
+
# - Scalable → use "that can handle growth"
|
| 86 |
+
# - Leverage → use "use"
|
| 87 |
+
# - Utilize → use "use"
|
| 88 |
+
# - Innovative → describe specific innovation
|
| 89 |
+
# - Optimal → use "best," "ideal," or "most effective"
|
| 90 |
+
# - Capability → use "ability," "feature," or "function"
|
| 91 |
+
# - Interface → use "connect," "work with," or describe interaction
|
| 92 |
+
# - Efficient → use "quick," "streamlined," or describe time savings
|
| 93 |
+
# - Paradigm → use "model," "approach," or "way of thinking"
|
| 94 |
+
# - Facilitate → use "help," "make easier," or "support"
|
| 95 |
+
# - Optimize → use "improve," "refine," or describe enhancement
|
| 96 |
+
|
| 97 |
+
# ### Specific Content Elements to Include When Relevant
|
| 98 |
+
|
| 99 |
+
# #### German Language Skills
|
| 100 |
+
|
| 101 |
+
# - Mention motivation for learning German and timeline goals
|
| 102 |
+
# - Address language barrier concerns proactively
|
| 103 |
+
|
| 104 |
+
# #### Technology Learning Appetite
|
| 105 |
+
|
| 106 |
+
# - Show willingness to learn new technologies (e.g., SAP, Vue.js, Ai, LLMs)
|
| 107 |
+
# - Highlight strong fundamentals that enable quick learning
|
| 108 |
+
# - Provide examples of past rapid skill acquisition
|
| 109 |
+
|
| 110 |
+
# #### Experience Quality Over Quantity
|
| 111 |
+
|
| 112 |
+
# - Emphasize depth and relevance of experience
|
| 113 |
+
# - Highlight specific achievements and impact
|
| 114 |
+
# - Show progression and growth in current role
|
| 115 |
+
|
| 116 |
+
# ## Output Rules
|
| 117 |
+
|
| 118 |
+
# 1. **Content Only:** Provide only the greeting and 3-paragraph body content - no headers, contact information, date, subject line, or closing signature
|
| 119 |
+
# 2. **Customization:** Every paragraph must be specifically tailored to the job description
|
| 120 |
+
# 3. **Authenticity:** Write like a real human - natural flow, simple language, avoid overly polished corporate speak
|
| 121 |
+
# 4. **Specificity:** Reference specific company details, role requirements, and technologies mentioned in the JD
|
| 122 |
+
|
| 123 |
+
# ## Quality Checklist
|
| 124 |
+
|
| 125 |
+
# Before finalizing, ensure the paragraphs meet these criteria:
|
| 126 |
+
|
| 127 |
+
# - [ ] Addresses the specific role and company mentioned in the job details by name
|
| 128 |
+
# - [ ] Highlights 2-3 most relevant experiences from the CV
|
| 129 |
+
# - [ ] Addresses any obvious skill gaps with learning commitment
|
| 130 |
+
# - [ ] Uses natural, varied language (avoiding the banned word list)
|
| 131 |
+
# - [ ] Uses simple, clear language that sounds human-written (not AI-generated)
|
| 132 |
+
# - [ ] Avoids overly sophisticated vocabulary or complex sentence structures
|
| 133 |
+
# - [ ] Includes a clear call to action
|
| 134 |
+
# - [ ] Stays within 3-4 paragraph limit
|
| 135 |
+
|
| 136 |
+
# Remember: Your goal is to write paragraphs that sounds like it was written by a real person with good English skills, not by AI. Keep it natural, straightforward, and genuine.
|
| 137 |
+
|
| 138 |
+
# """
|
| 139 |
+
|
| 140 |
+
|
| 141 |
+
# print(len(prompt),'length')
|
| 142 |
+
|
| 143 |
+
# messages = [
|
| 144 |
+
# {"role": "system", "content": prompt}
|
| 145 |
+
# ]
|
| 146 |
+
# # Apply chat template
|
| 147 |
+
# text = tokenizer.apply_chat_template(
|
| 148 |
+
# messages,
|
| 149 |
+
# tokenize=False,
|
| 150 |
+
# add_generation_prompt=True
|
| 151 |
+
# )
|
| 152 |
+
|
| 153 |
+
# # Tokenize input
|
| 154 |
+
# model_inputs = tokenizer(
|
| 155 |
+
# [text],
|
| 156 |
+
# return_tensors="pt",
|
| 157 |
+
# padding=True,
|
| 158 |
+
# truncation=True,
|
| 159 |
+
# max_length=16000
|
| 160 |
+
# ).to(model.device)
|
| 161 |
+
|
| 162 |
+
# # Generate with parameters optimized for markdown
|
| 163 |
+
# generated_ids = model.generate(
|
| 164 |
+
# **model_inputs,
|
| 165 |
+
# max_new_tokens=7192, # Increased for longer markdown content
|
| 166 |
+
# do_sample=True,
|
| 167 |
+
# temperature=0.8, # Slightly lower for more consistent formatting
|
| 168 |
+
# top_p=0.9,
|
| 169 |
+
# top_k=50,
|
| 170 |
+
# repetition_penalty=1.15, # Higher to avoid repetitive formatting
|
| 171 |
+
# eos_token_id=tokenizer.eos_token_id,
|
| 172 |
+
# pad_token_id=tokenizer.pad_token_id
|
| 173 |
+
# )
|
| 174 |
+
|
| 175 |
+
# generated_only_ids = generated_ids[:, model_inputs.input_ids.shape[1]:]
|
| 176 |
+
|
| 177 |
+
|
| 178 |
+
# # Decode output
|
| 179 |
+
# response = tokenizer.decode(
|
| 180 |
+
# generated_only_ids[0],
|
| 181 |
+
# skip_special_tokens=True
|
| 182 |
+
# )
|
| 183 |
+
|
| 184 |
+
# return response.strip()
|
| 185 |
+
|
| 186 |
+
|
| 187 |
+
|
| 188 |
+
# except Exception as e:
|
| 189 |
+
# return {"error": str(e)}
|
| 190 |
+
|
| 191 |
+
|
| 192 |
+
# async def job_details_alteration(job_details:str) -> str:
|
| 193 |
+
# try:
|
| 194 |
+
# prompt = f"""
|
| 195 |
+
# job_description: {job_details}
|
| 196 |
+
|
| 197 |
+
# Your task: Extract only the important parts and rewrite them clearly into 4 sections:
|
| 198 |
+
# - **Responsibilities** (what the candidate will do)
|
| 199 |
+
# - **Requirements / Qualifications** (skills, education, experience needed)
|
| 200 |
+
# - **Who You Are** (traits, mindset, culture fit)
|
| 201 |
+
# - **Preferred Candidate** (optional nice-to-have skills or experience)
|
| 202 |
+
|
| 203 |
+
# Rules:
|
| 204 |
+
# - Remove irrelevant parts such as "About Us", "Why Join Us", "Perks/Benefits", or generic company marketing.
|
| 205 |
+
# - Keep the output concise and professional.
|
| 206 |
+
# - If a section is not found, skip it. """
|
| 207 |
+
|
| 208 |
+
# messages = [
|
| 209 |
+
# {"role": "system", "content": prompt}
|
| 210 |
+
# ]
|
| 211 |
+
# text = tokenizer.apply_chat_template(
|
| 212 |
+
# messages,
|
| 213 |
+
# tokenize=False,
|
| 214 |
+
# add_generation_prompt=True
|
| 215 |
+
# )
|
| 216 |
+
# model_inputs = tokenizer(
|
| 217 |
+
# [text],
|
| 218 |
+
# return_tensors="pt",
|
| 219 |
+
# padding=True,
|
| 220 |
+
# truncation=True,
|
| 221 |
+
# max_length=2048
|
| 222 |
+
# ).to(model.device)
|
| 223 |
+
# generated_ids = model.generate(
|
| 224 |
+
# **model_inputs,
|
| 225 |
+
# max_new_tokens=8192,
|
| 226 |
+
# do_sample=True,
|
| 227 |
+
# temperature=0.5,
|
| 228 |
+
# top_p=0.9,
|
| 229 |
+
# top_k=50,
|
| 230 |
+
# repetition_penalty=1.15,
|
| 231 |
+
# eos_token_id=tokenizer.eos_token_id,
|
| 232 |
+
# pad_token_id=tokenizer.pad_token_id
|
| 233 |
+
# )
|
| 234 |
+
# generated_only_ids = generated_ids[:, model_inputs.input_ids.shape[1]:]
|
| 235 |
+
# response = tokenizer.decode(
|
| 236 |
+
# generated_only_ids[0],
|
| 237 |
+
# skip_special_tokens=True
|
| 238 |
+
# )
|
| 239 |
+
# return response.strip()
|
| 240 |
+
# except Exception as e:
|
| 241 |
+
# return {"error": str(e)}
|
| 242 |
+
|
| 243 |
+
|
| 244 |
+
async def coverLetterGenerativeAIBot(data: GenerateRequest, resume_text: str) -> str:
|
| 245 |
+
try:
|
| 246 |
|
| 247 |
+
user_content = f"""
|
| 248 |
+
You are a professional cover letter writer.
|
|
|
|
|
|
|
| 249 |
|
| 250 |
+
## Input
|
| 251 |
+
- **Job Title:** {data.job_title}
|
| 252 |
+
- **Job Description:** {data.job_details}
|
| 253 |
+
- **Resume:** {resume_text}
|
| 254 |
|
| 255 |
+
## Task
|
| 256 |
+
Write ONLY the four body paragraphs of a professional cover letter, following this structure:
|
| 257 |
|
| 258 |
+
1. **Paragraph 1**: Introduction where you're studying, and that you're applying for a student/intern position while studying.
|
| 259 |
+
2. **Paragraph 2**: Previous experience or academic projects, include 2 to 4 bullet points highlighting key projects or achievements.
|
| 260 |
+
3. **Paragraph 3**: Why you want to join this company what attracts you to them and what you can contribute.
|
| 261 |
+
4. **Paragraph 4**: Closing short and to the point, showing enthusiasm and availability.
|
| 262 |
+
|
| 263 |
+
### Language Requirements
|
| 264 |
+
|
| 265 |
+
- **Tone:** Professional and formal (German market prefers formal tone over conversational)
|
| 266 |
+
- **Length:** Concise and focused (3 paragraphs preferred for German market, maximum 4 paragraphs)
|
| 267 |
+
- **Total word count:** 250-400 words maximum
|
| 268 |
+
- **Per paragraph:** 70-100 words (3-4 sentences each)
|
| 269 |
+
- **Opening:** 50-70 words
|
| 270 |
+
- **Body:** 80-100 words
|
| 271 |
+
- **Closing:** 50-70 words
|
| 272 |
+
|
| 273 |
+
- **Avoid These Overused Words:**
|
| 274 |
+
- Robust → use "strong," "reliable," or "sturdy"
|
| 275 |
+
- Keen → use "eager," "interested," or "enthusiastic"
|
| 276 |
+
- Scalable → use "that can handle growth"
|
| 277 |
+
- Leverage → use "use"
|
| 278 |
+
- Utilize → use "use"
|
| 279 |
+
- Innovative → describe specific innovation
|
| 280 |
+
- Optimal → use "best," "ideal," or "most effective"
|
| 281 |
+
- Capability → use "ability," "feature," or "function"
|
| 282 |
+
- Interface → use "connect," "work with," or describe interaction
|
| 283 |
+
- Efficient → use "quick," "streamlined," or describe time savings
|
| 284 |
+
- Paradigm → use "model," "approach," or "way of thinking"
|
| 285 |
+
- Facilitate → use "help," "make easier," or "support"
|
| 286 |
+
- Optimize → use "improve," "refine," or describe enhancement
|
| 287 |
+
|
| 288 |
+
#### German Language Skills
|
| 289 |
|
| 290 |
+
- Mention motivation for learning German and timeline goals
|
| 291 |
+
- Address language barrier concerns proactively
|
| 292 |
|
| 293 |
+
#### Technology Learning Appetite
|
| 294 |
|
| 295 |
+
- Show willingness to learn new technologies (e.g., SAP, Vue.js, Ai, LLMs)
|
| 296 |
+
- Highlight strong fundamentals that enable quick learning
|
| 297 |
+
- Provide examples of past rapid skill acquisition
|
| 298 |
|
| 299 |
+
## Important:
|
| 300 |
+
- Do **not** include greetings (e.g., "Dear Hiring Manager")
|
| 301 |
+
- Do **not** include sign-offs (e.g., "Sincerely")
|
| 302 |
+
- Do **not** include headers or formatting
|
| 303 |
+
- Do **not** add number on paragraphs, also after each paragraph give a one line space gap then starts next paragraph
|
| 304 |
+
- Return **only** the four paragraphs
|
| 305 |
+
"""
|
| 306 |
|
| 307 |
+
openAiApiClient = OpenAI(api_key=os.getenv("OPENAI_API_KEY"))
|
| 308 |
+
|
| 309 |
messages = [
|
| 310 |
+
{"role": "system", "content": "You are a professional cover letter body paragraphs generator."},
|
| 311 |
+
{"role": "user", "content": user_content}
|
| 312 |
]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 313 |
|
| 314 |
+
# response = openAiApiClient.chat.completions.create(
|
| 315 |
+
# model="gpt-4-0613",
|
| 316 |
+
# messages=messages,
|
| 317 |
+
# max_tokens=6092
|
| 318 |
+
# )
|
| 319 |
+
# Choose model and max token limit
|
| 320 |
+
model = "gpt-4-0613"
|
| 321 |
+
max_total_tokens = 8192 # hard limit for gpt-4-0613
|
| 322 |
+
|
| 323 |
+
# Load tokenizer
|
| 324 |
+
encoding = tiktoken.encoding_for_model(model)
|
| 325 |
+
|
| 326 |
+
# Count tokens in the messages
|
| 327 |
+
def count_message_tokens(messages, encoding):
|
| 328 |
+
tokens_per_message = 3 # Every message object has a cost (role + content + metadata)
|
| 329 |
+
tokens_per_name = 1 # If 'name' is present
|
| 330 |
+
total_tokens = 0
|
| 331 |
+
|
| 332 |
+
for message in messages:
|
| 333 |
+
total_tokens += tokens_per_message
|
| 334 |
+
for key, value in message.items():
|
| 335 |
+
total_tokens += len(encoding.encode(value))
|
| 336 |
+
if key == "name":
|
| 337 |
+
total_tokens += tokens_per_name
|
| 338 |
+
total_tokens += 3 # Every reply is primed with <|start|>assistant<|message|>
|
| 339 |
+
return total_tokens
|
| 340 |
+
|
| 341 |
+
input_tokens = count_message_tokens(messages, encoding)
|
| 342 |
+
available_output_tokens = max_total_tokens - input_tokens
|
| 343 |
+
|
| 344 |
+
# Ensure we don't go negative
|
| 345 |
+
if available_output_tokens < 1:
|
| 346 |
+
raise ValueError("Input messages are too long. Reduce their size.")
|
| 347 |
+
|
| 348 |
+
# Call API
|
| 349 |
+
response = openAiApiClient.chat.completions.create(
|
| 350 |
+
model=model,
|
| 351 |
+
messages=messages,
|
| 352 |
+
max_tokens=available_output_tokens
|
| 353 |
)
|
| 354 |
|
| 355 |
+
ai_text = response.choices[0].message.content
|
| 356 |
+
|
| 357 |
+
return ai_text
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 358 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 359 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 360 |
except Exception as e:
|
| 361 |
+
print(f"Error: {e}")
|
| 362 |
+
return ""
|
app/services/hf_storage_service.py
CHANGED
|
@@ -18,11 +18,11 @@ class HuggingFaceStorageService:
|
|
| 18 |
raise ValueError("Folder must be 'resumes' or 'cover-letters'")
|
| 19 |
|
| 20 |
if filename is None:
|
| 21 |
-
filename = f"{uuid.uuid4().hex}
|
| 22 |
|
| 23 |
# Create a unique path with date
|
| 24 |
timestamp = datetime.now().strftime("%Y/%m/%d")
|
| 25 |
-
file_path = f"{folder}/{timestamp}/{filename}"
|
| 26 |
|
| 27 |
# Save bytes to temp file
|
| 28 |
with tempfile.NamedTemporaryFile(delete=False, suffix=os.path.splitext(filename)[1]) as temp_file:
|
|
|
|
| 18 |
raise ValueError("Folder must be 'resumes' or 'cover-letters'")
|
| 19 |
|
| 20 |
if filename is None:
|
| 21 |
+
filename = f"{uuid.uuid4().hex}"
|
| 22 |
|
| 23 |
# Create a unique path with date
|
| 24 |
timestamp = datetime.now().strftime("%Y/%m/%d")
|
| 25 |
+
file_path = f"{folder}/{timestamp}/{filename}.pdf"
|
| 26 |
|
| 27 |
# Save bytes to temp file
|
| 28 |
with tempfile.NamedTemporaryFile(delete=False, suffix=os.path.splitext(filename)[1]) as temp_file:
|
app/services/pdf_creator.py
CHANGED
|
@@ -1,7 +1,5 @@
|
|
| 1 |
from fpdf import FPDF
|
| 2 |
import os
|
| 3 |
-
from datetime import datetime
|
| 4 |
-
import io
|
| 5 |
from app.services.hf_storage_service import HuggingFaceStorageService
|
| 6 |
import unicodedata
|
| 7 |
import markdown2
|
|
@@ -61,9 +59,16 @@ def build_cover_letter_md(
|
|
| 61 |
job_title, company_name, generated_paragraphs
|
| 62 |
):
|
| 63 |
template = f"""
|
|
|
|
|
|
|
|
|
|
| 64 |
|
| 65 |
-
{
|
| 66 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 67 |
"""
|
| 68 |
return template
|
| 69 |
|
|
@@ -74,3 +79,5 @@ def convert_md_to_text(md_text: str) -> str:
|
|
| 74 |
html = markdown2.markdown(md_text)
|
| 75 |
plain_text = "".join(html.split("<")[0] if "<" in html else html for html in html.split(">"))
|
| 76 |
return plain_text
|
|
|
|
|
|
|
|
|
| 1 |
from fpdf import FPDF
|
| 2 |
import os
|
|
|
|
|
|
|
| 3 |
from app.services.hf_storage_service import HuggingFaceStorageService
|
| 4 |
import unicodedata
|
| 5 |
import markdown2
|
|
|
|
| 59 |
job_title, company_name, generated_paragraphs
|
| 60 |
):
|
| 61 |
template = f"""
|
| 62 |
+
{your_name}
|
| 63 |
+
{postal_code}, {city}
|
| 64 |
+
{email} | {phone}
|
| 65 |
|
| 66 |
+
Dear {company_name},
|
| 67 |
|
| 68 |
+
{generated_paragraphs}
|
| 69 |
+
|
| 70 |
+
Warm Regards,
|
| 71 |
+
{your_name}
|
| 72 |
"""
|
| 73 |
return template
|
| 74 |
|
|
|
|
| 79 |
html = markdown2.markdown(md_text)
|
| 80 |
plain_text = "".join(html.split("<")[0] if "<" in html else html for html in html.split(">"))
|
| 81 |
return plain_text
|
| 82 |
+
|
| 83 |
+
|
requirements.txt
CHANGED
|
@@ -9,4 +9,7 @@ safetensors
|
|
| 9 |
python-multipart
|
| 10 |
huggingface_hub
|
| 11 |
python-dotenv
|
| 12 |
-
markdown2
|
|
|
|
|
|
|
|
|
|
|
|
| 9 |
python-multipart
|
| 10 |
huggingface_hub
|
| 11 |
python-dotenv
|
| 12 |
+
markdown2
|
| 13 |
+
langchain
|
| 14 |
+
openai
|
| 15 |
+
tiktoken
|