Business Problem Statement: Master students at CRIOBE needed an accessible introduction to artificial intelligence and its applications for marine research. While they possessed strong backgrounds in biology and ecology, most lacked exposure to AI concepts and practical tools. The workshop bridged the gap between marine science theory and AI practice, demonstrating direct applications to coral reef monitoring and conservation research.

Workshop Format & Approach: Duration: 4 hours (2h lecture + 2h practical session)
Date: December 10, 2024
Format: In-person at CRIOBE Moorea research station
Participants: Master students in Marine Biology and Environmental Sciences

Curriculum Delivered:

Theoretical Foundation (10:30–12:30):

• AI History & Evolution: Comprehensive overview of artificial intelligence development (~45 minutes)
• Key Vocabulary & Concepts: Essential AI terminology and foundational concepts
• Myth-Busting: Addressing common misconceptions about AI capabilities and limitations
• Machine Learning Fundamentals: Introduction to core ML principles and applications (~1h15)

Practical Application Session (13:30–15:30):

• Generative AI Tools: Hands-on demonstrations of ChatGPT, Midjourney, and LaMa inpainting
• Discriminative AI Models: Practical experience with YOLO for object detection
• Marine Research Applications: Direct application of AI tools to coral reef imagery analysis
• Hands-on Exercise: Automated detection of quadrat grids in coral survey photographs

Quantifiable Outcomes:

• Complete Engagement: 100% of students successfully completed all practical exercises
• Technical Proficiency: Students gained first-hand experience with both generative and discriminative AI tools
• High Satisfaction: Average satisfaction score of 4.7/5 (collected informally)
• Academic Integration: Several students expressed strong interest in integrating AI methods into upcoming thesis projects
• Practical Skills: Students demonstrated ability to apply AI tools to real coral reef research scenarios

Academic Impact & Applications: The workshop introduced CRIOBE students to state-of-the-art AI techniques with direct relevance to coral reef research, providing them with a practical foundation for exploring AI in their academic work. Following the training, students began considering AI applications for:

• Coral Classification: Automated species identification in reef surveys
• Reef Monitoring Automation: Streamlined analysis of underwater photography
• Ecological Pattern Detection: AI-powered analysis of environmental data
• Research Efficiency: Accelerated data processing and analysis workflows

Long-term Research Value:

• Enhanced Research Capabilities: Students equipped with cutting-edge AI tools for marine research
• Academic Innovation: Integration of AI methods into marine biology thesis projects
• Research Quality: Improved accuracy and efficiency in coral reef monitoring and analysis
• Knowledge Transfer: Students positioned to advance AI applications in marine conservation

This workshop successfully demonstrated how specialized AI training can enhance academic research capabilities, bridging the gap between advanced technology and marine science applications while maintaining focus on practical implementation and research relevance.

Our coral monitoring solution leverages photo quadrat imagery collected from designated sites in French Polynesia to provide detailed, time-series assessments of coral reef coverage. By employing advanced computer vision and machine learning techniques, the system automatically detects and classifies coral species. The implementation involves taking high-resolution underwater pictures at specific coral sites, processing the collected images through an AI-powered analysis platform, and then display the results on a centralized dashboard.

Using a deep learning solution to count juvenile holothurians. Having a solution that is robust to changes in illumination and shadows requires a more advanced treatment than traditional thresholding techniques. An AI method popularized in cell counting in microbiology has been adapted to count holothurians and provide counting software.

Business Problem Statement: Many professionals in French Polynesia were curious about AI but lacked practical understanding of how to apply it in their daily work. Common challenges included misconceptions about AI capabilities, uncertainty about concrete business benefits, and a need for hands-on training with accessible tools to improve workplace efficiency and innovation. The workshop addressed the gap between AI awareness and practical implementation.

Workshop Format & Approach: Duration: 4-hour intensive workshop
Format: In-person at PÉPITES training center
Participants: Mixed group of professionals from various sectors (administration, business, education) — approximately 12–15 participants per session

Curriculum Delivered:

• AI Fundamentals: Introduction to AI concepts and evolution (non-technical overview)
• Vocabulary & Myth-Busting: Clarifying terminology and debunking common misconceptions about AI
• Practical Demonstrations: Hands-on experience with generative AI tools (ChatGPT, Midjourney)
• Local Applications: Exploration of AI solutions developed “for the fenua” (locally adapted)
• Interactive Exercises: Participants identified real-world applications relevant to their professional contexts
• Group Discussions: Collaborative reflection on integrating AI into individual workflows and organizational practices

Quantifiable Outcomes:

• Confidence Building: 90% of participants reported significantly improved confidence in using AI tools
• Practical Application: Participants identified at least 2 concrete use cases directly relevant to their daily work
• Satisfaction Rating: Post-workshop feedback averaged 4.6/5 across all sessions
• Implementation Success: Several participants implemented ChatGPT for document drafting, customer communication improvement, and task automation within one month
• Knowledge Retention: Participants demonstrated clear understanding of AI limitations and potential risks

Participant Feedback & Impact: Participants highlighted the clarity of explanations and the effective balance between theory and practical application. Many appreciated discovering AI in a non-intimidating environment and gained valuable understanding of both AI capabilities and limitations. The workshop successfully transformed abstract AI concepts into actionable workplace tools.

Long-term Business Value:

• Workforce Development: Enhanced AI literacy across diverse professional sectors in French Polynesia
• Practical Implementation: Direct application of AI tools in participant workflows within weeks
• Risk Awareness: Educated approach to AI adoption with clear understanding of limitations and ethical considerations
• Local Market Impact: Contributed to building AI competency in the French Polynesian professional ecosystem

This workshop demonstrated the effectiveness of hands-on, culturally relevant AI training for professional development, successfully bridging the gap between AI technology and practical workplace application.

Development of an AI-based automatic counting solution for microplastics in giant clams as part of Irène Godere’s PhD thesis (supervised by Nabila Gaertner-Mazouni), aimed at studying plastic contamination in French Polynesia

Accounting entry of invoices is a tedious task and many solutions exist on the market to help accelerate the process. Traditional image processing combined with rule-based algorithms do not cope well with the growing diversity of invoice templates. Hence, deep learning based solutions with continuous training are now the leading tehnologies as they can adapt to new templates provided that enough examplars are seen. In this work, a generic state of the art OCR service has been deployed along with a custom trained key information retrieval neural network model to infer the different accounting fields on the invoice.

Digitizing Single Line Diagrams (SLD) is a long and expensive process due to the large diversity of symbols and the high number of occurences. Recent advances in object detection and fine-grained image classification offer good solutions to detect and recognize electrical symbols from schemas automatically. A proprietary database was constructed to train and evaluate state of the art solutions for spotting and then recognizing each symbol.

Business Problem Statement: The client’s mission was to monitor public works tenders of a given city and detect overlaps between projects targeting the same roads. The objective was to identify cases where different utility providers (telecom companies, water utilities, etc.) planned to excavate the same road within short timeframes. By alerting providers of these overlaps, they could synchronize their interventions, avoiding repeated digging and refilling of roads, saving both time and money while reducing citizen inconvenience.

Diagnostic Process: Our 3-session diagnostic workshop included:

• Session 1: Current tender monitoring processes and document format analysis across different utility providers
• Session 2: AI technology evaluation for natural language processing and geospatial analysis of tender documents
• Session 3: Implementation strategy for automated overlap detection and provider alert system

Key Findings & Recommendations:

• Natural language processing (NLP) and information extraction identified as key technologies for parsing tender descriptions and extracting location, timeframes, and work types
• Geospatial AI and mapping techniques recommended to match tenders to specific road segments and detect temporal/geographic overlaps
• Example scenario: System could flag when a telecom operator plans underground cable replacement on a street in June while a water utility schedules pipe maintenance on the same street in July
• Rule-based alert system proposed to notify providers of collaboration opportunities and help city coordinate roadworks
• Implementation challenges: Variability in tender document formats, incomplete geolocation data, and need for city-wide data integration

Quantifiable Outcomes:

• Implementation investment: €200K–€350K budget estimates for pilot solution depending on geospatial integration complexity
• Efficiency improvements: 15–25% reduction in redundant roadworks through coordinated planning
• Cost savings: Significant cost savings for utility providers through synchronized interventions
• Time optimization: Several weeks saved per overlapping project through coordinated planning
• Citizen impact: Improved satisfaction due to fewer and shorter roadwork disruptions

Recommended Next Steps:

1. Phase 1: Build proof of concept to parse tenders and detect overlaps in one city district (6 months)
2. Phase 2: Integrate geospatial mapping capabilities to refine overlap detection accuracy
3. Phase 3: Scale solution to cover entire city tender database and create alerting dashboard for providers and city planners
4. Stakeholder engagement: Develop collaboration framework with utility providers for coordinated planning

This diagnostic provided the client with a comprehensive strategy for using AI to optimize city-wide infrastructure coordination, reducing redundant works and improving urban planning efficiency.

Recent advances in deep learning have show great potential for crowd counting from 2D images. While several solutions require dedicated cameras (2d or 3d) for high precision counting, recent methods leverage existing surveillance cameras and achieving decent performance around 90% accuracy. The severe occlusion in crowded spaces remains challenging but promising performance are obtained in somewhat controlled indoor environement. Density-based approaches as well as head detection approches have been evaluated on high view point scenes in airports and close view point in buses.

Time-lapses of construction sites provide a great way to monitor and supervise construction activities. Construction vehicule presence is a good indicator of the building stage and having statistics over time provide valuable insight for monitoring. While construction vehicule have only a few classes (dump truck, compactor, excavator, loader, mobile crane), their appearance changes a lot due to their articulated structure and many disctractors are present on the scene. A performance study as conducted to evaluate state of the art object detection solutions on this task.

Business Problem Statement: Conducted for a company specializing in home assistance with medical equipment, this diagnostic addressed the challenge of optimizing delivery routes for large numbers of devices (oxygen concentrators, beds, wheelchairs) dispatched to customers’ homes. The client needed to reduce delivery delays, improve planning for equipment maintenance and retrieval, and optimize their fleet operations across multiple service areas.

Diagnostic Process: Our 3-session diagnostic workshop included:

• Session 1: Current routing operations assessment and data extraction challenges from existing ERP systems
• Session 2: Optimization algorithm evaluation for delivery/pickup routing and demand forecasting capabilities
• Session 3: Implementation roadmap for predictive maintenance scheduling and real-time route optimization

Key Findings & Recommendations:

• Real-time routing optimization for equipment delivery and pickup identified as primary opportunity, reducing travel time significantly
• Demand forecasting models recommended to anticipate seasonal peaks (e.g., flu seasons) and optimize stock allocation across warehouses
• Predictive maintenance scheduling system proposed to prevent equipment breakdowns and improve customer satisfaction
• Key challenges identified: Data quality issues in historical logistics records, integration complexity with existing planning software, and staff adoption concerns
• ERP integration barrier: Existing data was not easily extractable from their current enterprise resource planning system

Quantifiable Outcomes:

• Travel efficiency gains: 20–30% reduction in daily travel distance, saving approximately 2–3 hours per driver per day
• Cost reduction potential: 10–15% reduction in fuel and logistics costs across the entire fleet
• Equipment utilization improvement: 20–25% better utilization of equipment stock, reducing idle equipment inventory
• Service reliability: Maintenance-related breakdowns projected to drop by 25–30%, improving overall service reliability
• Resource optimization: Better allocation of drivers and vehicles based on predictive demand patterns

Recommended Next Steps:

1. Phase 1: Pilot routing optimization system in one regional hub to validate improvements (3–6 months)
2. Phase 2: Roll out demand forecasting models using historical delivery and maintenance records
3. Phase 3: Deploy predictive maintenance system leveraging repair history and IoT sensor data on critical equipment
4. Integration strategy: Develop data extraction solutions for existing ERP system to enable AI implementation

This diagnostic provided the client with a comprehensive strategy for leveraging AI to optimize their home medical equipment logistics operations, with clear ROI projections and implementation priorities.

In order to design custom eartips, 3d scans of the ear and its canal are required. A solution based on CNNs has been developped in order to track ear landmarks during smartphone scaning. Inspired by facial landmarks research, a deep learning based solution was trained on a proprietary database to detect ear landmarks. The solution achieved high accuracy and has been deployed successfully.

Food volume estimation is an important challenge with many applications ranging from measuring calories in a diet management program to estimating the cooking time depending on the nature and thickness of the ingredients when following a recipe.

A study was conducted to experiment vision-based solutions and a prototype was developped in the form of an android app. From multiple pictures of the ingredients, the app is able to recover the thickness and surface of each ingredient with good precision.

Traditional and off-the-shelf OCR solutions could not read leather markings correctly due to various text orientations and special image texture. Hence, a custom OCR solution was developed and patented. As the Lead Tech in this project, I was designated as one of the inventors.

For more information refer to Neovision project page or directly to CTC’s project page

Business Problem Statement: The client wanted to explore the potential of AI for accounting workflows, driven by the rising interest in deep learning and document understanding technologies. The objective was to assess which accounting processes (invoice processing, reconciliation, reporting) could realistically be automated with AI, while providing concrete budget estimates for building invoice parsing software and other AI-enabled automations. The diagnostic also examined risk factors, feasibility constraints, and business areas with the strongest potential ROI.

Diagnostic Process: Our 3-session diagnostic workshop included:

• Session 1: Current accounting workflow mapping and data infrastructure assessment, focusing on invoice processing bottlenecks
• Session 2: AI technology evaluation for document processing, OCR capabilities, and NLP-based field extraction
• Session 3: Implementation roadmap development with budget scenarios and risk assessment framework

Key Findings & Recommendations:

• Invoice parsing using AI (OCR + NLP) was identified as the most promising entry point for automation
• Expense categorization and anomaly detection in financial reporting were highlighted as viable medium-term opportunities
• Reconciliation and compliance checking were identified as longer-term, higher complexity use cases requiring more advanced AI
• Key risks identified: data quality issues in existing invoices, change management challenges in finance teams, and uncertainty around vendor tool maturity
• Data annotation challenge: While many invoices were available, they were only partially annotated, requiring additional preparation for AI model training

Quantifiable Outcomes:

• Implementation budget scenarios: €150K–€250K for initial invoice parsing software deployment
• Expected efficiency gains: 50–70% reduction in manual data entry time across invoice processing workflows
• ROI timeline: Faster invoice turnaround and reduced processing overhead identified as primary value drivers
• Risk mitigation strategies: Decision framework provided to guide strategic AI investment choices

Recommended Next Steps:

1. Phase 1: Develop proof of concept for invoice parsing using a small subset of invoices (3–6 months)
2. Phase 2: Evaluate ROI and data quality improvements before scaling to other accounting processes
3. Phase 3: Gradually expand to expense categorization and anomaly detection once invoice automation proved business value
4. Decision framework: Structured approach provided to guide future AI investment priorities

This diagnostic equipped the client with realistic budget expectations, concrete ROI projections, and a phased implementation strategy for integrating AI into their accounting operations.