Shams Roumili

Sifflet is Europe’s leading Enterprise Data Quality & Observability platform. In this role, I : • Supported the full sales cycle: technical qualification, demos, discovery, scoping, and POCs. • Led enterprise proof-of-concepts and designed custom solutions using Sifflet’s API to secure technical wins against jointly defined success criteria. • Delivered post-sales onboarding and training to drive adoption and long-term customer success. • Built a customer support chatbot (Chainlit, Claude 4.5) delivering tailored data quality insights and product guidance. Technical stack: Snowflake, Databricks, dbt, Airflow, PowerBI, Looker, Chainlit

Leverage LLMs to extract and retrieve engine parts maintenance information from highly variable and unstructured document formats. Automatically update technical documents after the maintenance visit. Technical stack: Python, Pandas, streamlit, langchain, AWS Bedrock, Git, dvc

Straw-Berry reinvents the beverage experience with straws made from sugarcane fiber, which combine pleasure, robustness, innovation and durability. As COO, I drove the company from a buy/resell concept to a recognized player in sustainable hospitality solutions with its own production line and a strong market presence. Specifically, I : •Led procurement of raw materials and machinery, supply chain optimization, and R&D for product development •Implemented tools to support the business : ERP (Odoo), WMS, scrapers (Python, selenium) •Developed multi-channel sales strategies (Google Ads, newsletters, direct outreach, phone, trade shows, LinkedIn, website, samples, and sales agents) •Successfully pitched to investors and incubators (NeuillyLab) & secured funding 🤝 MamaShelter 🤝 Accor 🤝 Best Western 🤝 Pernod Ricard 🤝 Paris Society 🤝 Hôtel George V 🤝 Camping Paradis

In this position, I was in charge of sizing, developing and presenting AI-based tools to assist key functions of Corporate Investment Banking within BNP. I had the opportunity to work with a wide range of stakeholders, both technical and non-technical. • Developed an internal tool to automatically detect, qualify and escalate market abuse alerts (ex : Front Running, Wash trading). • Developed Anti Money Laundering tools to reduce the workload of AML analysts by 42%. • Built a classifier for Sanction, Embargo and Terrorism Alerts to reduce the workload of analysts by 60%. Technical stack: Python, pandas, sklearn, PySpark, shap, MLflow, ElasticSearch, Oracle APEX, Oracle Exadata, Docker, Kubernetes, Apache Kafka, Ansible, Git

Designed and implemented road obstacle detection methods for autonomous vehicles. • Combined state of the art machine learning with data augmentation and computer vision methods (edge detection, superpixels, saliency) to counter failure cases: achieved 0.51 Average Precision • Attained results interpretability by generating activation heat maps and ablation studies on the architecture The solutions were sold to a major automotive group. Technical Stack: Python, Pytorch, MLflow, AWS (S3 & EC2), OpenCV, Bitbucket

-Banking transaction analysis (Python, pandas, sklearn) : data preprocessing (clean, remove outliers, create quantitative features from exogenous data, normalize), clustering. Identification of 3 client clusters and suggestion of personalized offers for each cluster. -Development of modules for data crawling (SQL, NoSQL, Trello, Slack), data exploration (PCA, clustering, regression) and vizualisation (pivot table, correlation matrix, graph). -Fleet management : analysis of driving data of a major vehicle rental firm. Identification of driving behaviours that lead to accidents. -Creation of a job offer/CV matcher: pdf parsing, named entity recognition, embedding of CV and offers, matching by cosine similarity.