Mohamed Mammar

Développement d'une plateforme Next-Gen Agentic pour les internes.

- Developed autonomous AI agents and an Agent Orchestrator (LangChain, LangGraph, AWS Bedrock AgentCore) enabling generation of complex workflows from simple prompts. - Designed a drag-and-drop platform for creating end-to-end WhatsApp chatbot conversational workflows, increasing the number of demos presented to clients daily by 10x. - Integrated Agent-to-Agent (A2A) protocols for collaboration across multi-agent systems, with LangSmith for tracing, debugging, and observability. - Developed MCPs around Meta services (WhatsApp, Catalogs, etc.) assisting with over 50% of our team’s tasks. - Implemented a Long-Term Memory (LTM) management system, increasing session quality via a dynamic cosine similarity threshold (in %) determined by a supervisor Agent. - Built an agentic web scraping pipeline via BS4 & Playwright, automating spreadsheet updates for products scraped from the client’s website and their publication to an operational WhatsApp catalog, increasing new client onboarding speed by 25%. - Designed and deployed an autonomous WhatsApp Agent on AWS Lambda and MongoDB to promote client brands’ products. - Developed an API automating prospecting, WhatsApp messaging campaigns, and CRM updates. - Integrated an optimized CI/CD pipeline for the monorepo, with selective infrastructure deployments. - Resolved a 6-month-old performance bottleneck in a router Lambda by migrating the code from Python to JavaScript to stream the response to the WhatsApp webhook, eliminating 50% of redundant requests caused by its unanswered retries. - Standardized project architecture via an N-Layered Architecture approach.

Momentum Technologies is an IT consulting firm specializing in data, with over 20 years of experience. It supports Quebec organizations in managing their data and offers AI solutions to problems in various fields. The company has nearly 200 experts, with an estimated impact on more than 12 million users, and manages over 25,000,000 terabytes of data. Among their clients: Ramq and Ciusss. I joined the Data Scientist team at Momentum Technologies in 2022 to work on client and internal projects. As a Data Scientist, my responsibilities included: - Participation in AI initiative committees: market research, proposing innovative solutions, improving responses to calls for tenders. - Design and implementation of ETL pipelines, feature engineering, data exploration, visualization, and modeling. - Development of a chatbot with an LLM-RAG pipeline, capable of answering over 90% of questions from job seekers and recruiters. - Development of advanced AI fraud detection algorithms such as GBM, SVM, DBSCAN. - Implementation and maintenance of models developed on AWS services (S3, Lambda, SageMaker, Bedrock). I contributed to the following developments: - Momentum Technologies' revenue increased by +10%. - Extension of several contracts with large groups. - Automation of responses to various and diverse requests.

AGROLAB is a laboratory group active in the fields of agriculture and food safety/security with over 10 years of experience, specializing in microscopic image analysis for quality control and research. My responsibilities included: - Integration and processing of microscopic image data via ETL pipelines. - Development of Computer Vision pipelines using YOLO models to detect impurities (pollen, residues) in microscopic images of honey and food. - Fine-tuning of Vision Transformers (ViT-Base) and EfficientNet-B7 models to classify honey and pollen types and identify potential structural anomalies in protein cells. - Development of a U-Net architecture integrating attention mechanisms, specifically designed for precise segmentation of microstructures in images of food samples. - Model optimization through the application of classic and synthetic Data Augmentation techniques (StyleGAN-3) to improve the robustness and performance of segmentation. - Participation in the deployment of models on edge devices via ONNX Runtime and TensorRT, with optimization via pruning and quantization for connected microscopes, which drastically reduced inference latency. In less than a year, I was able to make major contributions to the development of time-consuming task automation while reducing their initial cost. I contributed to the following improvements: - AGROLAB's revenue increased by +15%. - 3 new clients (large groups) and 5 others. - Extension of several contracts with large groups. - Acceleration of the laboratory's scientific research, which increased by 25%.