📄 Skill Matrix Filler App

An intelligent document automation tool that extracts structured data from resumes and dynamically fills DOCX templates — including complex Skill/Qualification matrices — using local LLMs (via Ollama) and ChromaDB-powered RAG.

---

🚀 Features

• ✅ Upload and process resumes in .pdf, .docx, or .doc formats
• ✅ Analyze template structure using LLM to infer expected fields
• ✅ Extract structured resume data (skills, experience, certifications)
• ✅ Dynamically fill any resume-based template using extracted data
• ✅ Special support for Skill/Qualification Matrix templates
• ✅ Preview filled templates as HTML and download as .docx
• ✅ Ask questions about the resume/template using a simple chat interface
• ✅ All data is embedded and queried using ChromaDB
• ✅ Runs fully offline using Ollama + Mistral (no OpenAI API needed)

---

🧠 How It Works – 10 Steps

1. Upload Resume and Template: Users upload a resume and a DOCX template via the web interface.

2. Save & Track Session: Files are saved with timestamps, and a unique session_id is generated.

3. Parse Resume to Structured JSON: Text is extracted and sent to a local LLM to output fields like name, experience, education, etc.

4. Analyze Template Schema: A second LLM analyzes the empty template to understand its layout, sections, and repeatable blocks.

5. Embed Content with ChromaDB: All text content and parsed metadata are stored as vector embeddings in ChromaDB for RAG-based retrieval.

6. Display Extracted Data: The parsed resume data is shown in JSON form in the UI for user verification.

7. Trigger Template Filling: The backend uses the resume data + template schema to build LLM instructions for dynamic field filling.

8. Fill the Template Dynamically: With python-docx, the app fills in the fields, paragraphs, and tables using LLM-generated content.

9. Save and Serve Filled DOCX: The filled document is saved under filled_docs/ and made downloadable via a backend endpoint.

10. Preview + Q&A: Users can preview the DOCX as HTML and ask questions via a mini chat powered by the LLM and ChromaDB.

---

📂 Project Structure

Skill_Matrix_App/
├── app.py                         # Main Flask backend
├── core/
│   ├── llama_runner.py            # Handles Ollama LLM prompts
│   ├── resume_parser.py           # Extracts structured data from resume
│   ├── template_analyzer.py       # Infers structure of blank templates
│   ├── template_processor.py      # Fills templates dynamically
│   └── chroma_service.py          # Handles embedding/indexing in ChromaDB
├── utils/
│   └── file_utils.py              # File reading, cleaning, chunking
├── static/
│   └── index.html                 # Frontend UI
├── uploads/                       # Uploaded files
├── filled_docs/                   # Output files
└── app.log                        # Runtime logs

---

🛠️ Requirements

• Python 3.10+
• Ollama installed and running locally
• Model pulled: mistral:7b-instruct-v0.3-q4_K_M
• pip install -r requirements.txt (typical: flask, sentence-transformers, chromadb, python-docx, mammoth, fitz, ollama, etc.)

---

🧪 Running the App

# Start the Ollama server first
ollama serve

# Ensure model is pulled
ollama pull mistral:7b-instruct-v0.3-q4_K_M

# Then run the Flask app
python app.py

# Open in browser
http://localhost:8008

---

✨ Live Demo Features

• Upload .pdf, .docx, or .doc resumes

• Upload empty or semi-filled templates

• Click "Upload & Ingest" to parse + embed

• Click "Fill Template" to get the final DOCX

• Preview as HTML and download

• Ask questions like:

  • “What certifications does the candidate hold?”
  • “List all companies worked at in the last 5 years”

---