Grégoire Mariot

Ownership of 6 AI solutions in production. I own the product vision, define and prioritize the roadmap, write functional specifications, and continuously arbitrate between business urgency and technical feasibility. I coordinate IT teams (ServiceNow, DataMiner) and business stakeholders (operations, NOC, supervision, technical management). I lead sprint reviews, steering committees, and am responsible for the value delivered at each iteration.

Power BI is my central tool for management on this mission. I use it to monitor operational KPIs of deployed AI solutions (automation rate, processed volumes, error rate), for model performance analysis, and for decision-making reporting.

- Automatic Validation of Prevention Plans— Autonomous AI system for matching site access requests with active prevention plans. Power BI monitoring of automation rate and residual cases.

- Automated Customer Communication in ServiceNow— Automatic generation of communications compliant with internal rules, integrated into ServiceNow. Significant reduction in workload for NOC teams. Power BI dashboard monitoring of generated communication volumes and quality.

- Automation of Critical Email Processing (Energy Flow)— Automatic classification of 4,000+ annual power outage alerts. Intelligent parsing, structured extraction, automatic creation of ServiceNow tickets. Zero manual operation for standard cases.

- AI-Augmented Data Analysis & Power BI— Automatic generation of visualizations and dashboards. Combination of natural language data exploration with Power BI dashboards for decision-making.

Stack:Python, Power BI, ServiceNow, DataMiner LangGraph/LangChain, LangSmith, Anthropic Claude, OpenAI, Mixtral (fine-tuning), SQL, REST APIs

Client:Sanofi Aventis R&D

Sector:Pharmaceutical / R&D

Conditions:Hybrid (Vitry-sur-Seine)

Sanofi's R&D scientific equipment inventory (53,024 pieces of equipment, 7 sites, 4 countries) was fragmented into 342 categories and 989 sub-categories, lacking a coherent global view. Traditional methods (Excel, 3 people, several weeks) had failed.

Product Owner for the AI classification and harmonization project of the equipment inventory. I own the product vision and project roadmap. Scoping the need with stakeholders (IT, lab managers, R&D business units), prioritizing functional modules, coordinating multi-site and multi-country (France, Germany, USA, Belgium). Leading 3-4 human-in-the-loop validation cycles with senior scientific experts. Arbitrating between scope coverage and classification quality.

Public presentation of the Use Case at the Adopt IA 2025 trade show.

Power BI is used throughout the project for tracking classification progress by site and country, monitoring confidence scores by category, analyzing discrepancies between validation iterations, and for executive reporting presented to program sponsors. Power BI dashboards allow stakeholders to visualize classification progress in real-time and identify areas requiring human attention.

Multi-model generative AI classification (OpenAI + Anthropic Claude), graph matching with confidence scoring, automated quality control.

- 53,024 equipment classified across 7 sites in 4 countries

- Reduction from 342 to 22 categories (-93%)

- Reduction from 989 to 387 sub-categories (-61%)

- Accuracy rate: 80-90%

Stack:Python, Power BI, OpenAI, Anthropic Claude, LangChain, LangSmith, SQL

Role:AI & Data Consultant

Sector:Industry / Equipment Distribution

Kubota had a catalog of 1.5 million product sheets with insufficient raw content for e-commerce: missing or poor descriptions, absent SEO metadata, unexploited images.

Design and deployment of a large-scale automated product sheet enrichment AI solution. Project management from the audit phase to production.

- Automated Product Sheet Enrichment— Transformation of 1.5 million raw product sheets into optimized marketing content through automatic image analysis and AI text generation. Each sheet enriched with commercial descriptions, technical specifications, and SEO metadata.

- AI-Powered SEO Optimization— 35% improvement in organic traffic through automatic optimization of 600,000+ product sheets (titles, descriptions, image alt tags). Results measured on actual e-commerce site traffic.

- 1.5 million enriched sheets

- +35% measured organic traffic

- 600,000+ SEO-optimized sheets

Stack:Python, Generative AI (image analysis + text generation), AI SEO