Mehdi Mehdi

# Project 1: Event-Driven PDF Ingestion & Structured Extraction Pipeline: - Enabled seamless collaboration between legal and data teams by automating litigation data transfer, tracking, and analysis, significantly reducing manual processing & operational friction. - Built event-driven ingestion pipeline on GCP detecting PDF uploads via Google Drive API and orchestrated with Cloud Functions. - Implemented LLM + LlamaIndex-based OCR and data extraction, storing structured outputs in BigQuery tables. - Used Vertex AI LLMs to generate concise document summaries for analytics and downstream consumption. - Integrated Langfuse for LLM monitoring, prompt tracing, and performance optimization. # Project 2: AI-Powered Profession Classification & Insurance Risk Assessment System: - Reduced insurance underwriting preparation time through real-time profession matching with transparent confidence scoring and match-tier classification. - Developed 3-tier hybrid search system using FastAPI and FAISS, combining exact matching, fuzzy matching, and semantic search for 1,186 French professions across 15 sectors. - Automated NAF-profession association matching 1,186 sector professions to 7,967 official codes using - OpenAI embeddings & LLM models with LangChain, and Pydantic schema validation with confidence thresholds. - Built dual-interface solution with RESTful API (~150ms response) and StreamlitUI, implementing Langfuse tracing for comprehensive LLM observability and token cost monitoring.

Project 1: HR Automation Platform: - Engineered end-to-end web application to streamline recruitment, reducing candidate selection time by 4hrs/week. - Developed anonymization pipeline for GDPR compliance and hybrid OCR system for reliable PDF data extraction. - Implemented new LLM-based CV classification (x6 accuracy improvement) with temporal depreciation weighting to prioritize recent experience. - Built CRM integration for dynamic talent pool dashboards and automated candidate outreach (LinkedIn messaging assistant). # Project 2: "CV-Mission request Matcher" Consultant Matching System: - Created mission-matching solution scoring profiles on technical skills (0-5 scale) and time￾weighted experience. - Designed dual-model evaluation: Skills gap analysis + recency-weighted professional history scoring. - Developed reporting module highlighting: Matching/missing competencies, relevant experiences, visual ranking dashboards. - Reduced sales team response time for technical proposals through automated top-3 candidate identification.

# Project 3 "Juritrack" Legal document analysis & insight extraction (Clustering & Topic modeling) - Updated a Law firm client project and introduced new features, leading to increased revenue. - Performed data engineering tasks using GCP and its Vision API, optimizing data extraction and processing workflows. - Utilized advanced NLP & LLM tools to enhance data analysis accuracy and processing efficiency. - Increased client productivity by approximately 45% through contributions to the project.