llm_integration.md

LLM Integration

The application relies heavily on Large Language Models (LLMs) for understanding job descriptions and generating human-like text for resumes and cover letters.

LLM Manager (src/libs/llm_manager.py)

This module handles the abstraction layer for different AI providers.

AI Model Adapter Pattern

The AIAdapter class acts as a factory, instantiating the correct model class based on the configuration (LLM_MODEL_TYPE).

Supported Providers:

• OpenAI (OpenAIModel)
• Claude (ClaudeModel)
• Ollama (OllamaModel) - for local inference
• Gemini (GeminiModel)
• HuggingFace (HuggingFaceModel)
• Perplexity (PerplexityModel)

GPTAnswerer

The GPTAnswerer class is a high-level service that uses the configured LLM to answer specific questions related to the resume or job application.

Key Features:

• answer_question_textual_wide_range: Determines which section of the resume (e.g., Experience, Education) is relevant to a question and uses an appropriate prompt chain to generate an answer.
• is_job_suitable: Analyzes the job description against the resume to calculate a suitability score.
• summarize_job_description: Compresses long job descriptions into concise summaries.

Logging (LLMLogger)

All LLM requests and responses are logged to open_ai_calls.json for debugging and cost tracking. It captures:

• Model Name
• Token Usage (Input/Output/Total)
• Estimated Cost
• Prompts and Replies

Prompt Engineering

Prompts are stored in src/libs/llm/prompts.py (referenced in llm_manager.py). The application uses LangChain templates to structure these prompts dynamically with input variables like {resume_section} or {job_description}.