Zakaria HAMANE

- Led and mentored a team of data scientists in building production AI systems, providing technical guidance on architecture decisions, code reviews, and best practices for LLM-based applications, while fostering a collaborative environment focused on continuous learning and innovation. - Built a GraphRAG system that extracts entities and relationships from documents using LLMs, stores them in Neo4j knowledge graphs, and enables natural language querying through automated Cypher generation combined with vector similarity search. - Designed a distributed task processing engine with Celery and Redis featuring intelligent dependency resolution, automatic parallel/sequential execution detection, and smart rate-limit handling across multiple specialized worker queues. - Implemented flexible document processing pipelines supporting multiple parsing tools (PyPDF, Docling, LlamaParse), various chunking strategies, and advanced query processing with rewriting, expansion, and decomposition capabilities, all managed through version-controlled prompts in Langfuse. - Integrated Neo4j, ChromaDB, MinIO, Elasticsearch, and PostgreSQL with optimized connection pooling and async operations, deployed via Docker with comprehensive monitoring.

- Developed TalkerAI, an AI-powered Voice Assistant for Sales that conducts real-time conversations using LLMs (GPT-4, Gemini, Groq), Speech-to-Text (Deepgram), and Text-to-Speech (ElevenLabs), featuring multi-agent function calling for domain-specific tasks, Twilio integration for call handling, distributed system architecture for concurrent conversations, multilingual support (English/French), and comprehensive monitoring systems. - Led and mentored a team of 5 data scientists, providing strategic guidance, technical direction, and code reviews, resulting in improved project outcomes and elevated team performance. Delivered targeted training and constructive feedback, fostering a culture of continuous learning and innovation. - Developed a distributed task processing system using Celery and Redis (which uses LLMs to generate R&D Tax Credits reports), implemented a web server using FastAPI with endpoints for generating reports, retrieving task status, retrying, regenerating, and cancelling tasks, and integrated WebSocket for real-time task status updates. Utilized Docker for containerization, ensuring easy deployment and scalability, with a setup including the main application, Celery workers, Celery beat for task scheduling, and Flower for monitoring Celery tasks. - Built an LLMOps system for monitoring LLMs and Retrieval-Augmented Generation (RAG) systems and fine-tuning LLM models on our internal reports.

- Information Retrieval from Documents using LLM Models: Developed and implemented a system for information retrieval from a large corpus of documents using Language Model models (LLM). Applied techniques such as BERT, GPT-3, or similar models to extract relevant information and improve search capabilities. This project aimed to enhance the efficiency and accuracy of document search and retrieval processes within the organization.

• Successfully executed Record Linkage on various data sources including Crunchbase, Google Patents, FDA, Clinical Trials, Open Alex, and LinkedIn to link records of people and organizations from those data sources using (NLP, similarity techniques). • Developed and deployed a startups Recommender System for investors, resulting in increased efficiency and effectiveness in identifying promising startups to recommend to investors. • Implemented MLOps using MLFlow, streamlining the machine learning model development and deployment process. • Utilized Time Series Aggregation techniques for patents, scientific papers, and life science investment data, providing valuable insights for investors. • Contributed to the construction of a comprehensive knowledge Graph from linked data sources, enhancing the platform's ability to connect (funds, people, patents etc.). • Trained and deployed Graph Neural Network model using GraphSAGE, resulting in improved performance and accuracy dependent projects (e.g. Companies Similarity, Recommender System). • Worked as a Lead Data Scientist, provided Mentoring, Training, Technical Direction, Code Review during projects and feedback for my team.

• Development of an algorithm based on Natural language processing (𝐍𝐋𝐏) to automatically 𝐢𝐝𝐞𝐧𝐭𝐢𝐟𝐲 the right 𝐩𝐚𝐭𝐢𝐞𝐧𝐭𝐬 for the 𝐜𝐥𝐢𝐧𝐢𝐜𝐚𝐥 𝐭𝐫𝐢𝐚𝐥𝐬. • Researched, prototyped (from research papers), built features and optimized(hyper-parameter tuning) the state-of-the-art machine learning and deep learning techniques like 𝐒𝐕𝐌, 𝐋𝐨𝐠𝐢𝐬𝐭𝐢𝐜 𝐑𝐞𝐠𝐫𝐞𝐬𝐬𝐢𝐨𝐧, 𝐑𝐚𝐧𝐝𝐨𝐦 𝐅𝐨𝐫𝐞𝐬𝐭 𝐫𝐞𝐠𝐫𝐞𝐬𝐬𝐢𝐨𝐧, 𝐋𝐒𝐓𝐌, 𝐂𝐍𝐍 etc., using 𝐒𝐜𝐢𝐤𝐢𝐭-𝐋𝐞𝐚𝐫𝐧, 𝐊𝐞𝐫𝐚𝐬, 𝐓𝐞𝐧𝐬𝐨𝐫𝐟𝐥𝐨𝐰 on CPU/GPU environments for 𝐌𝐞𝐝𝐢𝐜𝐚𝐥 𝐓𝐞𝐱𝐭-𝐂𝐥𝐚𝐬𝐬𝐢𝐟𝐢𝐜𝐚𝐭𝐢𝐨𝐧. • Digging deep into millions of 𝐩𝐚𝐭𝐢𝐞𝐧𝐭𝐬 𝐝𝐚𝐭𝐚 𝐫𝐞𝐜𝐨𝐫𝐝𝐬 and performing data exploratory analysis using interactive 𝐉𝐮𝐩𝐲𝐭𝐞𝐫 𝐧𝐨𝐭𝐞𝐛𝐨𝐨𝐤𝐬 (𝐩𝐚𝐧𝐝𝐚𝐬/𝐧𝐮𝐦𝐩𝐲/𝐦𝐚𝐭𝐩𝐥𝐨𝐭𝐥𝐢𝐛) and presented the findings to the managers. • Lead the design and implemented container-based architecture in Cl/CD processes using 𝐆𝐢𝐭, 𝐃𝐨𝐜𝐤𝐞𝐫 𝐚𝐧𝐝 𝐊𝐮𝐛𝐞𝐫𝐧𝐞𝐭𝐞𝐬 and build 𝐃𝐚𝐭𝐚𝐎𝐩𝐬, 𝐌𝐋𝐎𝐩𝐬 at scale. • Extensively worked and enabled 𝐌𝐋𝐟𝐥𝐨𝐰 to manage the 𝐌𝐋 𝐥𝐢𝐟𝐞𝐜𝐲𝐜𝐥𝐞, including experimentation, reproducibility, deployment, and a central model registry. • Development (𝐆𝐀𝐍) algorithms to generate images human skin to solve the lack of Data problem, and using this data to 𝐂𝐥𝐚𝐬𝐬𝐢𝐟𝐲 𝐒𝐤𝐢𝐧 𝐂𝐚𝐧𝐜𝐞𝐫𝐬 types. • Developing a 𝐰𝐞𝐛 𝐚𝐩𝐩𝐥𝐢𝐜𝐚𝐭𝐢𝐨𝐧 containing interactive maps and graphs using 𝐏𝐥𝐨𝐭𝐥𝐲 and 𝐒𝐭𝐫𝐞𝐚𝐦𝐥𝐢𝐭. • 𝐏𝐫𝐨𝐯𝐢𝐝𝐞𝐝 mentoring, Trainings and 𝐓𝐞𝐜𝐡𝐧𝐢𝐜𝐚𝐥 𝐃𝐢𝐫𝐞𝐜𝐭𝐢𝐨𝐧 of Machine Learning and the process of Data Science projects to the cross-functional team • Contribute to 𝐜𝐨𝐝𝐞 𝐫𝐞𝐯𝐢𝐞𝐰𝐬 during projects and provide feedback • Managed and trained Interns on there end of studies projects -Applying 1) 𝐂𝐍𝐍-𝐑𝐍𝐍 architecture 2) 𝐘𝐨𝐥𝐨 pretrained model for 𝐒𝐤𝐢𝐧 𝐂𝐚𝐧𝐜𝐞𝐫 𝐂𝐥𝐚𝐬𝐬𝐢𝐟𝐢𝐜𝐚𝐭𝐢𝐨𝐧.

• Setting up a named entities recognition (𝐍𝐄𝐑) to extract information from "𝐦𝐞𝐝𝐢𝐜𝐚𝐥 𝐭𝐞𝐱𝐭𝐬" (prescriptions, clinical notes, etc.). Technologies: (𝐍𝐋𝐓𝐊, 𝐒𝐩𝐚𝐜𝐲, 𝐏𝐲𝐓𝐨𝐫𝐜𝐡) • Implementation of an 𝐎𝐂𝐑 solution based on Machine Learning for medical documents (prescription, clinical note, etc.). Technologies: (𝐎𝐂𝐑 , 𝐘𝐎𝐋𝐎, 𝐂𝐍𝐍, 𝐏𝐲𝐓𝐨𝐫𝐜𝐡) • Applied various transfer-learning techniques using pre-trained word-embeddings like 𝐆𝐥𝐨𝐯𝐞, 𝐟𝐚𝐬𝐭𝐓𝐞𝐱𝐭, 𝐆𝐥𝐨𝐯𝐞, 𝐁𝐄𝐑𝐓, 𝐄𝐋𝐦𝐨 for text similarity tasks. • Implementation of an 𝐍𝐋𝐏 based 𝐒𝐞𝐚𝐫𝐜𝐡 𝐞𝐧𝐠𝐢𝐧𝐞 that can understand users search queries for an electronic medical records management system. Technologies: (𝐍𝐋𝐓𝐊, 𝐒𝐩𝐚𝐜𝐲, 𝐄𝐥𝐚𝐬𝐭𝐢𝐜𝐬𝐞𝐚𝐫𝐜𝐡, 𝐩𝐲𝐭𝐡𝐨𝐧) • Conceptualized and implemented the vision, maturity model and roadmap for Data Science

• Implementation of a 𝐂𝐡𝐚𝐭𝐁𝐨𝐭 using 𝐑𝐚𝐬𝐚 for e-commerce platforms. • Implementation of a machine learning model for the detection of 𝐅𝐢𝐧𝐚𝐧𝐜𝐢𝐚𝐥 𝐅𝐫𝐚𝐮𝐝 and 𝐂𝐫𝐞𝐝𝐢𝐭 𝐂𝐚𝐫𝐝 𝐅𝐫𝐚𝐮𝐝 • Model development for optimizing 𝐏𝐫𝐨𝐝𝐮𝐜𝐭𝐬 𝐑𝐞𝐜𝐨𝐦𝐦𝐞𝐧𝐝𝐚𝐭𝐢𝐨𝐧 𝐒𝐲𝐬𝐭𝐞𝐦 for e-commerce platforms • Implementation of a 𝐂𝐫𝐞𝐝𝐢𝐭 𝐒𝐜𝐨𝐫𝐢𝐧𝐠 model for banks 𝐓𝐞𝐜𝐡𝐧𝐨𝐥𝐨𝐠𝐢𝐞𝐬: NLTK, Spacy, RASA, scikit-learn, python, Git

Business Intelligence Department: • Created and presented models for 𝐏𝐫𝐞𝐝𝐢𝐜𝐭𝐢𝐯𝐞 𝐌𝐚𝐢𝐧𝐭𝐞𝐧𝐚𝐧𝐜𝐞. Achieved 20% better returns vs historical performance using 𝐒𝐕𝐌, 𝐋𝐨𝐠𝐢𝐬𝐭𝐢𝐜 𝐑𝐞𝐠𝐫𝐞𝐬𝐬𝐢𝐨𝐧, 𝐑𝐚𝐧𝐝𝐨𝐦 𝐅𝐨𝐫𝐞𝐬𝐭 𝐫𝐞𝐠𝐫𝐞𝐬𝐬𝐢𝐨𝐧. Technologies: 𝐬𝐜𝐢𝐤𝐢𝐭-𝐥𝐞𝐚𝐫𝐧, 𝐩𝐲𝐭𝐡𝐨𝐧, 𝐆𝐢𝐭 • Setting up a predictive maintenance information system with Oracle BI and ODI. • Processing, cleaning and verification of Datasets

Created and presented models for Predictive Maintenance. Achieved 20% better accuracy vs historical performance using SVM, Logistic Regression, and Random Forest. Technologies: (Scikit-Learn, Python, Git) Setting up a predictive maintenance information system with Oracle BI and ODI. Processing, cleaning, and verification of Datasets.

Discovery of the Industry and Commerce sectors

End of study dissertation on "The determinants of decision-making"

• Machine Learning: Classification, Feature Engineering, Supervised Learning (decision trees, random forest, logistic regression etc.), Unsupervised Learning (clustering, dimensionality reduction etc.) optimization (gradient descent and variants) • Mathematics and statistics: Statistical modeling, Bayesian inference, laws of probabilities. • Functional skills: e-health, Banking, Insurance, e-commerce and Chemical Industry • Project Management: Agile Scrum, CRISP-Data Science