🚀 My Career Advisor

🛠️ Technologies

• Language: Python 3.10+
• AI Framework: LlamaIndex (RAG Orchestration)
• LLM Engine: GPT4All (Local Model Execution)
• Model: Llama 3.2 1B-Instruct (Quantized GGUF)
• Embeddings: HuggingFace bge-small-en-v1.5
• Interface: Streamlit
• API: Adzuna Job Search API

---

🌟 Features (What you can do)

• Secure Upload: Drag and drop your .pdf or .docx resume safely.
• Instant Analysis: Get a professional breakdown of your resume's strengths and weaknesses.
• Rating System: Receive scores on content quality, keywords, and market relevance.
• Live Job Matching: Automatically search for active job vacancies that match your profile.
• Data Privacy: Experience AI without an internet connection (except for job searching).

---

⌨️ Keyboard Shortcuts (Streamlit Standard)

• R: Rerun the application.
• C: Clear the cache (useful if you change the model file).
• Enter: Confirm file upload or trigger the analysis button when focused.

---

🏗️ The Creation Process

The development followed a "Privacy-First" philosophy.

1. Environment Setup: Created a virtual environment to manage complex AI dependencies.
2. Model Selection: Tested different GGUF models to find the perfect balance between speed and reasoning (Llama 3.2 1B).
3. Bridge Building: Developed a CustomLLM adapter class to connect the GPT4All local engine to the LlamaIndex framework.
4. RAG Implementation: Configured a vector store to index resume data for precise context retrieval.
5. API Integration: Connected the Adzuna service to bridge the gap between "AI analysis" and "Real-world action."

---

📝 What I Learned

• LLM Adapters: How to write custom classes to make incompatible libraries work together.
• Resource Management: Using @st.cache_resource to keep large models in RAM, preventing system lag.
• Prompt Engineering: How to guide small 1B models to stay focused on specific tasks without "hallucinating."
• RAG Architecture: The power of giving an AI a specific "memory" (the resume) rather than relying on its general knowledge.

---

📈 How it can be Improved

• Cover Letter Generation: Add a feature to draft a personalized cover letter for the found jobs.
• Multi-language Support: Expand the prompt templates to analyze resumes in Portuguese and Spanish.
• Database Integration: Allow users to save their analysis history locally using SQLite.
• Better Models: Testing Llama 3.2 3B or Mistral models for even deeper reasoning if hardware allows.

---

🚀 How to Run

1. Clone & Enter:

git clone https://github.com/JoyceAcacioPedro/My-Career.git

2. Install Requirements:

pip install streamlit llama-index gpt4all requests docx2txt llama-index-embeddings-huggingface

3. Model Path: Open My_Career.py and ensure the model_path matches your local GPT4All folder.
4. Run:

streamlit run My_Career.py

---

📺 Demo Video

https://youtu.be/KZlMncvWR2k

"Watch how My Career Advisor transforms a raw resume into a targeted career strategy in seconds."

---

Developed with 🧠 by Joyce Pedro