From research to reality.

Driving digital transformation through advanced AI and specialised software engineering. Five domain-agnostic engines you can drop into a real business, each one a side door from a peer-reviewed paper.

1. Live demo

   /01

   Hybrid Explainable Recommender

   e-commerce · cross-domain · Wordpress

   A hybrid explainable recommendation system that delivers personalised suggestions with clear, evidence-based reasoning behind every result.

   Our Hybrid Explainable Recommendation System is designed to deliver not only highly relevant recommendations, but also clear and meaningful explanations behind each decision.

   In many systems, recommendations are presented as opaque outputs — users see what is suggested, but not why. This solution addresses that gap by combining advanced recommendation modelling with built-in explainability, making every result both accurate and interpretable.

   What It Does

   The system analyses multiple layers of user and product signals to generate personalised recommendations, including:

   • Historical interaction behaviour
   • Purchase and conversion patterns
   • Collaborative signals across similar users

   Each result is accompanied by a short, natural-language explanation grounded in real data — derived directly from the evidence used during ranking.

   Why It Matters

   Understanding why something is recommended increases trust, improves user experience, and supports better decision-making. The system ensures that:

   • Recommendations remain consistent and justifiable
   • Explanations are aligned with actual model behaviour
   • Outputs can be validated and trusted

   Key Capabilities

   • Personalised recommendations based on user behaviour and contextual signals
   • Built-in explainability with evidence-based reasoning
   • Hybrid modelling combining historical, behavioural, and collaborative data
   • Natural-language output with human-readable explanations
   • Reusable, domain-agnostic architecture

   How It Works

   The system uses a hybrid recommendation model to rank items based on multiple signal sources. An integrated language model translates structured evidence into short explanations, ensuring every recommendation is both data-driven and understandable.

   Designed for Adaptation

   This system serves as a foundational layer adaptable into project-specific solutions. Its modular design allows integration into different platforms while preserving both recommendation performance and explainability.

   • Recommender Systems
   • Explainable AI
   • Hybrid Modelling
   • Natural-Language Output
   • Domain-Agnostic

2. Live demo

   /02

   Multimodal Emotion Recognition

   customer experience · workplace

   Voice prosody, facial micro-expressions and spoken content fused into a unified emotional profile — real-time, privacy-aware, federated-ready.

   Our Multimodal Emotion Recognition system reads the full signal a person sends: how they sound, how their face moves, and what they say. Instead of relying on a single channel, it fuses audio, video and text into one coherent emotional state — with both category and intensity.

   The architecture is designed for sensitive contexts: inference can run locally on the edge, models are federated-ready so personal data never has to leave the device, and the same engine works across customer-facing and workplace deployments.

   What It Does

   • Analyses voice tone, rhythm and prosody
   • Reads facial expressions and micro-expressions
   • Processes spoken content and contextual metadata
   • Predicts both emotion category and intensity

   Why It Matters

   Improves customer satisfaction by adapting to emotional context in real time. Detects stress, discomfort and disengagement early — before they escalate. Supports employee well-being via continuous, opt-in sentiment monitoring.

   Key Capabilities

   • Multimodal detection across audio, video and text
   • Real-time inference suitable for live interactions
   • Federated learning ready — train without centralising data
   • Privacy-aware design with on-device processing options
   • Configurable emotion taxonomies for different domains

   How It Works

   Three specialised encoders extract signals from voice, face and text in parallel. A fusion layer combines them into a unified emotional embedding, which a downstream head translates into category + intensity predictions. The whole pipeline runs on commodity hardware with low latency.

   Designed for Adaptation

   Drop the engine into a contact-centre, a retail experience, an HR well-being programme or a clinical study — the same core adapts via lightweight fine-tuning. The system is the multimodal foundation behind our work in EARS and ships with privacy controls aligned with GDPR and the AI Act.

   • Multimodal Detection
   • Real-time Inference
   • Federated Learning Ready
   • Privacy-Aware
   • Audio · Video · Text

3. Live demo

   /03

   ERP Virtual Assistant

   Cegid PHC · Sage · Cegid Primavera

   An intelligent conversational interface that simplifies how businesses interact with their management systems — invoices, reports, forecasts and answers, without the menus.

   Our ERP Virtual Assistant turns the ERP from a menu-driven tool into a conversational partner. Instead of navigating layers of forms and reports, users simply ask — and the system delivers documents, summaries, forecasts and answers grounded in their own business data.

   The assistant connects directly to Cegid PHC, Sage, Cegid Primavera and other ERPs, blending natural language understanding with domain-specific business logic so that every reply is both contextual and operationally accurate.

   What It Does

   • Generates invoices & financial documents automatically
   • Produces periodic reports and performance summaries
   • Delivers AI-powered forecasts from historical data
   • Answers business queries in natural language

   Why It Matters

   Reduces time spent on repetitive admin. Makes business data accessible to non-technical users. Enables faster, data-informed decisions. Provides proactive insights instead of reactive reporting.

   Key Capabilities

   • NLP layer over the ERP — talk to your business in plain language
   • Automated document generation (invoices, orders, statements)
   • AI forecasting trained on the client's historical operations
   • Multi-source data aggregation across modules and systems
   • Custom ERP API connectors — Cegid PHC, Sage, Cegid Primavera and beyond

   How It Works

   A natural-language layer interprets user intent, maps it to ERP-native operations, and orchestrates the right modules to answer or act. Forecasts and summaries are produced by domain models trained on the client's own historical data, so output stays specific and accountable.

   Designed for Adaptation

   Built as a modular layer on top of the existing ERP, it extends without disrupting. New connectors, new business rules and new data sources can be added incrementally — the assistant evolves alongside the operation.

   • NLP over ERP
   • Automated Documents
   • AI Forecasting
   • Multi-source Aggregation
   • Custom ERP API

4. Live demo

   /04

   Autonomous Agent Systems

   automation · orchestration

   AI-powered systems that execute tasks independently, make informed decisions, and coordinate operations with minimal human intervention.

   Our Autonomous Agent Systems turn isolated automations into a coordinated, goal-driven workforce. Each agent owns a slice of the operation — planning its own actions, calling the right tools and APIs, and adjusting based on feedback — so the whole pipeline keeps moving without a human at every step.

   Agents collaborate through a shared orchestration layer that tracks state, handles handoffs and resolves conflicts. The result is a system that operates more like a team than a script: resilient to surprise, capable of branching, and observable end-to-end.

   What It Does

   • Plans, executes and adjusts actions on feedback
   • Operates across multiple systems and APIs
   • Handles sequential and conditional task chains
   • Learns and improves through contextual feedback

   Why It Matters

   Adapts dynamically to context without manual intervention. Integrates multiple systems into unified workflows. Acts proactively rather than reactively. Reduces operational overhead and human error.

   Key Capabilities

   • Autonomous execution of long-running, multi-step tasks
   • API integration with internal and external services
   • Multi-step workflows with branching and recovery
   • Adaptive learning from outcomes and human feedback
   • Self-managing — agents allocate work amongst themselves

   How It Works

   A central orchestrator decomposes the goal into tasks and assigns them to specialised agents. Each agent reasons over its own tools, calls APIs, and reports back. The orchestrator keeps a shared memory of state and outcomes, enabling agents to coordinate, retry intelligently, and learn from previous runs.

   Designed for Adaptation

   Built as a modular framework: new agents, new tools, new domains can be added without rewriting the core. The same engine underpins automations across customer service, operations and back-office processes — and is reused inside our work on EARS and ELFMO.

   • Autonomous Execution
   • API Integration
   • Multi-Step Workflows
   • Adaptive Learning
   • Self-Managing

5. Live demo

   /05

   Retopology Converter

   3D pipelines · games · industrial viz

   Transforms high-poly models into optimised low-poly meshes while preserving essential detail and visual fidelity. Clean, quad-based topology suitable for animation and real-time applications.

   Our Retopology Converter takes dense, scan-derived or sculpted geometry and rebuilds it as clean, animation-ready topology — automatically. The result preserves silhouette and surface detail while replacing chaotic triangulation with predictable quad flow.

   The system was designed to live inside production pipelines: studios can drop in raw assets and get back meshes that are immediately usable for rigging, baking, real-time rendering or industrial simulation — without weeks of manual cleanup.

   What It Does

   • Reduces mesh complexity automatically
   • Preserves core geometry and shape
   • Maintains visual fidelity end-to-end
   • Generates clean quad-based topology

   Why It Matters

   Cuts production time and manual effort. Ensures consistent topology quality across an entire pipeline. Enables faster iteration and scaling for game development, film and industrial visualisation studios.

   Key Capabilities

   • Fully automated retopology — no manual quad-drawing required
   • Quad-based output suitable for animation and subdivision
   • Geometry & silhouette preservation across LODs
   • Real-time engine ready — Unreal, Unity, web 3D out-of-the-box
   • Batch processing for asset libraries and scan archives

   How It Works

   The pipeline analyses the input mesh's curvature, feature lines and topology flow, then generates a new quad-dominant surface aligned to those features. UVs and normals are transferred from the source so the converted asset retains its original look while running orders of magnitude lighter.

   Designed for Adaptation

   The converter is modular: studios plug it into existing DCC workflows (Blender, Maya, Houdini) or call it as a service from their own asset pipelines. The same engine underpins our work in EXPAI (industrial 3D) and is reused across consumer and enterprise projects.

   • Automated Retopology
   • Quad-Based Output
   • Geometry Preservation
   • Real-time Engine Ready

/ Expert Services

FTP Academy

Visit Academy ↗

An innovative learning platform designed to empower the professionals of tomorrow. High-quality training programs across a wide range of technology fields, with practical learning, continuous development and tailored learning paths.

Open to candidates between 18 and 40 years old. No strict academic prerequisites for beginner courses. What matters: genuine interest in technology, motivation to learn and evolve.

• Practical Learning
• Tailored Paths
• DGERT Certified
• Mentorship

Appstudio

FTP's dedicated environment for designing, prototyping and delivering custom business applications at speed. Modular components and modern development practices translate complex business requirements into functional digital products.

From internal tools to customer portals and mobile apps. We combine reusable building blocks with tailored business logic so every application fits operations and scales as they evolve.

• Internal tools & dashboards Operational apps that streamline daily work.
• Automation bots & notifications Smart agents that monitor data and act.

How we work

1. Discovery We sit with your team to understand the process, the pain points and the goal.
2. Prototype We turn the idea into a working prototype quickly so you can react to something real.
3. Build & iterate We refine the solution with continuous feedback until it fits like a glove.
4. Deploy & support We ship it, train your team and stay close to evolve as your business changes.