Wacim Belahcel

• Lead the architecture and delivery of GenAI and Document AI solutions for banking and enterprise operations, from discovery and technical design to production integration.

• Led a GenAI pipeline on Microsoft Foundry using Microsoft Agent Framework to extract and reconcile source information, validate business rules and automatically prefill PDF forms with human-in-the-loop controls.

• Led a GenAI workflow for comparing Word documents, aligning sections and clauses to identify additions, deletions and semantic changes with traceable evidence for reviewers.

• Built document classification and structured extraction pipelines combining OCR, GPT-4o and multimodal LLMs, Pydantic schemas, prompt engineering and deterministic validation.

• Defined reusable components, APIs, evaluation datasets and observability paterns; coordinated engineers and stakeholders around quality, scalability, security and delivery. OCR, APIs, LLM evaluation

• Led the development and production deployment of multimodal document understanding models for identity documents and insurance workflows: page and document-type classification, orientation correction, OCR, key information extraction and field validation.

• Developed custom deep-learning pipelines with multimodal Transformers and Computer Vision / NLP techniques for document recognition, tables and forms.

• Built scalable customer-feedback analytics on Azure Databricks using Spark, MLflow and reproducible ML pipelines.

• Mentored interns on table KIE and NLP segmentation; drove technical watch and engineering practices across Document AI, Computer Vision, NLP and MLOps.

Tech: Azure, Databricks, Spark, MLflow, PyTorch, Hugging Face, OpenCV, OCR, Transformers, Hydra, Git

• Developed NLP opinion-mining datasets and models; researched GANs and built a CNN-RNN image-captioning prototype.