Finished writer model

configuration.py

@@ -18,43 +18,58 @@ SUMMARIZER_MODEL = "claude-3-5-haiku-20241022"
 # - Qwen2.5-Coder-14B-Instruct works OK, but is not great at markdown formatting
 #   and tends to get some details wrong.
 # - Claude-3-5-Haiku is the best model for this task so far.
-AGENT_MODEL = "claude-3-5-haiku-20241022"
-
-# Will be used for tasked related to writing the resume such as selecting
-# the best GitHub repositories, writing the resume content, etc.
 WRITER_MODEL = "claude-3-5-haiku-20241022"
 
-AGENT_INSTRUCTIONS = """
+WRITER_INSTRUCTIONS = """
 You are an AI agent responsible for writing a resume based on the provided context. Your task is to generate a well-structured and professional resume that highlights the user's skills, experiences, and achievements.
-You will receive two pieces of JSON structured context: a job call and a LinkedIn profile. 
-
-LINKEDIN PROFILE EXAMPLE
-
-"structured_text": {
-    "sections": {
-      "contact_info": "Contact details",
-      "summary": "Personal summary statement",
-      "skills": "Skills list",
-      "experience": "List of work experiences",
-      "education": "List of degrees",
-      "other sections": "Any other relevant sections from LinkedIn profile"
-    },
+You will receive three pieces of JSON structured context: a job call, a LinkedIn resume and a list of relevant projects. Each of these will be formatted as follows:
+
+JOB CALL FORMAT
+
+{
+  "job_title": "Position",
+  "company_description": "Company or organization information",
+  "job_description": "Description of role and responsibilities",
+  "key_skills": "List of required sills",
+  "tools_technologies": "List of necessary tools and technologies",
+  "experience_level": "Prior experience necessary",
+  "education_requirements": "Desired education level"
 }
 
-JOB CALL EXAMPLE
+LINKEDIN RESUME FORMAT
 
-'Job title': 'Position title',
-'Company description': 'Description of employer',
-'Job description': 'Job description summary',
-'Key skills': 'Required skills list',
-'Experience level': 'Required experience',
-'Education requirements': 'Required education level or degree'
+{
+  "contact_info": "Applicant contact information",
+  "certifications": "Licenses and certifications",
+  "summary": "Applicant personal statement",
+  "experience": "Applicant professional experience",
+  "education": "Applicant education and degrees"
+}
 
+PROJECT LIST FORMAT
+
+{
+  "projects": [
+    {
+      "title": "Repository 1 title",
+      "description": "Repository 1 project description",
+      "technologies": "List of tools and technologies",
+      "link": "URL"
+    },
+    {
+      "title": "Repository 2 title",
+      "description": "Repository 2 project description",
+      "technologies": "List of tools and technologies",
+      "link": "URL"
+    },
+  ]
+}
 
 Use this information to create a comprehensive resume that emphasizes the match between the provided linkedin profile and the job call. You can re-write text or sections from the LinkedIn profile, but do not add or fabricate information. Everything in the resume should be based on the provided context. The resume should include the following sections:
 - Contact Information
 - Summary
 - Skills
+- Projects
 - Work Experience
 - Education
 
@@ -83,4 +98,15 @@ Prioritize more recent and active repositories that demonstrate the user's skill
 
 Respond with only this list of repository titles, without any additional text or explanation.
 
+"""
+
+PROJECTS_SECTION_PROMPT = """
+You are an AI agent responsible for writing the projects section of a resume based on selected GitHub repositories. Your task is to generate a well-structured and professional description of the projects that highlights the user's skills, contributions, and achievements.
+You will receive a list of repository titles and a job call. Use this information to create a comprehensive projects section that emphasizes the match between the provided repositories and the job call. You can re-write text or sections from the repositories, but do not add or fabricate information. 
+Everything in the projects section should be based on the provided context. Format your response as a JSON object with the following fields:
+- 'projects': A list of dictionaries, each containing:
+  - 'title': The title of the project
+  - 'description': A brief description of the project, including the user's role and contributions
+  - 'technologies': A list of technologies used in the project
+  - 'link': A link to the project repository
 """

functions/gradio.py

@@ -118,7 +118,6 @@ def process_inputs(
             _ = write_resume(linkedin_resume, github_repositories, job_post)
 
         except Exception as e:
-            result += f"\n❌ Resume generation failed: {str(e)}\n"
             logger.error("Resume generation failed: %s", str(e))
     else:
         logger.warning("Resume generation skipped - content missing")

functions/writer_agent.py

@@ -5,13 +5,12 @@ import json
 import logging
 import os
 from openai import OpenAI
-from smolagents import OpenAIServerModel, CodeAgent
 from configuration import (
     INFERENCE_URL,
-    AGENT_MODEL,
-    AGENT_INSTRUCTIONS,
+    WRITER_INSTRUCTIONS,
     WRITER_MODEL,
-    REPO_SELECTION_PROMPT
+    REPO_SELECTION_PROMPT,
+    PROJECTS_SECTION_PROMPT
 )
 
 
@@ -36,63 +35,67 @@ def write_resume(linkedin_resume: dict, github_repositories: list, job_call: dic
 
     logger.info("Selecting relevant GitHub repositories based on job call")
     project_repos = _choose_repositories(github_repositories, job_call)
-    print("project_repos:", project_repos)
 
-        # model = OpenAIServerModel(
-        #     model_id=AGENT_MODEL,
-        #     api_base=INFERENCE_URL,
-        #     api_key=os.environ.get("API_KEY"),
-        # )
+    logger.info("Writing projects section of the resume")
+    projects = _write_projects_section(project_repos, job_call)
 
-        # agent = CodeAgent(
-        #     model=model,
-        #     tools=[],
-        #     additional_authorized_imports=['json', 'pandas'],
-        #     name="writer_agent",
-        #     verbosity_level=1,
-        #     max_steps=20,
-        #     planning_interval=5
-        # )
 
-        # # Prepare instructions - combine default with user instructions and job summary
-        # instructions = AGENT_INSTRUCTIONS
+# Let the model select the most relevant repositories based on the job call
+    client = OpenAI(
+        base_url=INFERENCE_URL,
+        api_key=os.environ.get("API_KEY", "dummy-key-for-testing")
+    )
 
-        # if job_summary is not None:
-        #     instructions += f"\n\nJob Requirements and Details:\n{json.dumps(job_summary)}"
-        #     logger.info("Added job summary to agent prompt")
+    prompt = f'JOB CALL\n{job_call}\nLINKEDIN RESUME\n{linkedin_resume}\nPROJECTS\n{projects}'
 
-        # if user_instructions and user_instructions.strip():
 
-        #     instructions += f"\n\nAdditional user instructions:\n{user_instructions.strip()}"
-        #     logger.info("Added user instructions to agent prompt")
+    messages = [
+        {
+            'role': 'system',
+            'content': WRITER_INSTRUCTIONS
+        },
+        {
+            'role': 'user',
+            'content': prompt
+        }
+    ]
 
-        # submitted_answer = agent.run(
-        #     instructions + '\n' + json.dumps(content['structured_text']),
-        # )
+    completion_args = {
+        'model': WRITER_MODEL,
+        'messages': messages,
+    }
 
-        # logger.info("submitted_answer: %s", submitted_answer)
+    try:
+        response = client.chat.completions.create(**completion_args)
 
-        # # Create data directory if it doesn't exist
-        # data_dir = 'data'
+    except Exception as e:
+        response = None
+        logger.error('Error during job summarization API call: %s', e)
 
-        # if not os.path.exists(data_dir):
+    if response is not None:
+        response = response.choices[0].message.content
 
-        #     os.makedirs(data_dir)
-        #     logger.info("Created data directory: %s", data_dir)
+    # Create data directory if it doesn't exist
+    data_dir = 'data'
 
-        # # Save the resume to resume.md in the data directory
-        # resume_file_path = os.path.join(data_dir, 'resume.md')
+    if not os.path.exists(data_dir):
 
-        # try:
-        #     with open(resume_file_path, 'w', encoding='utf-8') as f:
-        #         f.write(submitted_answer)
+        os.makedirs(data_dir)
+        logger.info("Created data directory: %s", data_dir)
 
-        #     logger.info("Resume saved to: %s", resume_file_path)
+    # Save the resume to resume.md in the data directory
+    resume_file_path = os.path.join(data_dir, 'resume.md')
 
-        # except Exception as e:
-        #     logger.error("Failed to save resume to file: %s", e)
+    try:
+        with open(resume_file_path, 'w', encoding='utf-8') as f:
+            f.write(response)
+
+        logger.info("Resume saved to: %s", resume_file_path)
+
+    except Exception as e:
+        logger.error("Failed to save resume to file: %s", e)
 
-    return project_repos
+    return response
 
 
 def _choose_repositories(github_repositories: list, job_call: dict) -> list:
@@ -109,7 +112,6 @@ def _choose_repositories(github_repositories: list, job_call: dict) -> list:
 
     logger = logging.getLogger(f'{__name__}._choose_repositories')
 
-
     # Create a new repo list without the full README text - this way we can save on input tokens
     # by only sending the model the repo metadata, title, description, topics, etc.
     repo_data = [
@@ -149,19 +151,59 @@ def _choose_repositories(github_repositories: list, job_call: dict) -> list:
     if response is not None:
         response = response.choices[0].message.content
         response = ast.literal_eval(response)
-        print(f'Selected repositories {type(response)}: {response}')
 
     # Now use the repository selection response to filter the repositories
-    selected_repos = []
+    selected_repos = [
+        repo for repo in github_repositories if repo['name'] in response
+    ]
+
+    return selected_repos
 
-    for repo in github_repositories:
-        print(repo['name'])
 
-        if repo['name'] in response:
-            selected_repos.append(repo)
+def _write_projects_section(project_repos: list, job_call: dict) -> str:
+    """
+    Write the projects section of the resume based on selected GitHub repositories.
 
-    # selected_repos = [
-    #     repo for repo in github_repositories if repo['name'] in response
-    # ]
+    Args:
+        project_repos (list): List of relevant GitHub repositories.
+        job_call (dict): Extracted/summarized job call information.
 
-    return selected_repos
+    Returns:
+        str: Formatted projects section for the resume.
+    """
+
+    logger = logging.getLogger(f'{__name__}._write_projects_section')
+
+    # Let the model select the most relevant repositories based on the job call
+    client = OpenAI(
+        base_url=INFERENCE_URL,
+        api_key=os.environ.get("API_KEY", "dummy-key-for-testing")
+    )
+
+    messages = [
+        {
+            'role': 'system',
+            'content': f'{PROJECTS_SECTION_PROMPT}'
+        },
+        {
+            'role': 'user',
+            'content': f'JOB CALL\n{json.dumps(job_call)}\n\nREPOSITORIES\n{json.dumps(project_repos)}'
+        }
+    ]
+
+    completion_args = {
+        'model': WRITER_MODEL,
+        'messages': messages,
+    }
+
+    try:
+        response = client.chat.completions.create(**completion_args)
+
+    except Exception as e:
+        response = None
+        logger.error('Error during job summarization API call: %s', e)
+
+    if response is not None:
+        response = response.choices[0].message.content
+
+    return response