suntux57420/stx-cv-parser repository overview

⁠STX CV Parser

STX CV Parser is an AI-powered microservice dedicated to automated candidate analysis (CVs and cover letters) against a job description.
It is developed as part of the Krokmou project, a modular on-premise AI platform designed to industrialize generative AI use cases in a controlled and sovereign way.

This service is intended as a decision-support tool for HR and recruitment teams, not as a replacement for human judgment.

⁠Key Features

• Automated analysis of CVs and cover letters
• Semantic matching between candidates and job descriptions
• OCR support for PDF documents (via external OCR microservice)
• Embeddings-based similarity search (Qdrant)
• Multi-criteria candidate scoring
• Structured HR-friendly summaries (strengths, weaknesses, recommendations)
• Interview question generation
• Interactive chat to explore and explain analysis results
• Designed for Docker and Kubernetes (K3s) deployments

⁠How It Works

STX CV Parser operates as a multi-step analysis pipeline:

1. Document ingestion
   CVs, cover letters and job descriptions are uploaded (PDF or text).

2. OCR processing
   PDF documents are converted to text using an external OCR microservice.

3. Text normalization
   Content is cleaned and normalized using a Large Language Model (LLM) to ensure consistent comparison across heterogeneous profiles.

4. Vectorization & indexing
   Documents are transformed into embeddings and stored in a Qdrant vector database.

5. Semantic comparison & scoring
   Candidate profiles are compared to the job description using a combination of:

   • vector similarity
   • rule-based criteria
   • qualitative LLM evaluation

6. Result generation
   The service produces structured summaries, scores, SWOT-style insights and interview recommendations.

7. Conversational exploration
   Users can interact with analysis results through a contextual chat interface.

⁠Architecture Overview

STX CV Parser relies on the following components:

• FastAPI — REST API and orchestration layer
• PostgreSQL — relational storage for analyses and metadata
• Qdrant — vector database for semantic search
• External OCR service — PDF to text conversion
• LLM API — semantic analysis and synthesis
• Embeddings API — vector generation

All components are configurable via environment variables.

⁠Running the Container

⁠Prerequisites

• Docker (or Docker Compose / Kubernetes)
• A running PostgreSQL instance
• A running Qdrant instance
• Access to:
  • an OCR microservice
  • an LLM API
  • an embeddings API

⁠Example: Run with Docker

docker run -d \
  --name stx-cv-parser \
  -p 8000:8000 \
  -e DATABASE_URL="postgresql+psycopg://user:password@postgres:5432/dbname" \
  -e QDRANT_URL="http://qdrant:6333" \
  -e OCR_URL="http://ocr-service:8010/api/convert" \
  -e OCR_WAIT="true" \
  -e OCR_TIMEOUT="180" \
  -e LLM_BASE_URL="https://api.llm.ev1/v1" \
  -e LLM_API_KEY="<YOUR_LLM_API_KEY>" \
  -e LLM_MODEL="gpt-4o-mini" \
  -e EMBED_BASE_URL="https://api.llm.ev1/v1" \
  -e EMBED_API_KEY="<YOUR_EMBED_API_KEY>" \
  -e EMBED_MODEL="mistral-embed" \
  -e EMBED_DIM="1024" \
  suntux57420/stx-cv-parser:latest

Once started, the service is available at:

http://localhost:8000

⁠Kubernetes / K3s

STX CV Parser is designed to run natively on Kubernetes (including K3s), using:

• ConfigMaps for non-sensitive configuration
• Secrets for API keys
• Persistent volumes for PostgreSQL and Qdrant

Refer to the project documentation for full Kubernetes manifests and deployment examples.

⁠Project Information

• Author: Guillaume Evain
• Contact: [email protected]⁠
• Git repository: https://gitlab.ev1.fr/ev1/stx-cv-parser/⁠
• Docker Hub: https://hub.docker.com/r/suntux57420/stx-cv-parser⁠
• Documentation: https://wiki.ev1.fr⁠
• Part of: Krokmou Project

⁠Philosophy

STX CV Parser focuses on:

• Transparency and explainability
• Data sovereignty and on-premise readiness
• Modular, reusable AI components
• Human-centered decision support

It demonstrates how generative AI can be integrated responsibly into sensitive business workflows.