FRAGMENTED
Distributed information
Knowledge is spread across websites, files, repositories, departments, and disconnected business systems.
Secure enterprise knowledge assistants
SAVIA: specialized AI assistants built around governed organizational knowledge
SAVIA is a modular platform developed by Pangeanic for creating, managing, and operating specialized virtual assistants. It connects advanced search, question answering, summarization, and controlled content generation with the knowledge, permissions, and security policies of each organization.
INREIA/2024/112 Official project file number
2025 to 2026 Development from January 2025 to May 2026
€399,048.85 Published project investment
Pangeanic Project coordinator and technology developer
ValgrAI Subcontracted research collaboration
The enterprise knowledge problem
Organizations have more information than ever, but much of it remains difficult to retrieve and control
Enterprise knowledge is distributed across websites, shared drives, internal documents, technical repositories, databases, and departmental systems. Employees often know that an answer exists without knowing where to find it.
General-purpose assistants can make this problem worse when they answer without reliable access to approved sources, ignore permission boundaries, or send sensitive information to external systems.
SAVIA addresses this gap by connecting specialized assistants to governed collections of organizational knowledge, while preserving administrative control over access, behavior, traceability, and deployment.
ACCESS
Uneven permissions
Different employees, teams, and external users require different levels of access to the same information.
TRUST
Uncontrolled answers
Generic assistants may produce plausible responses without grounding them in approved organizational sources.
AUDIT
Limited traceability
Organizations need to understand which users asked what, which sources were used, and how responses were generated.
The SAVIA platform
One environment for creating and operating multiple specialized assistants
SAVIA provides a centralized architecture for connecting knowledge sources, configuring assistants, assigning skills and permissions, routing requests, generating responses, and measuring usage.
INGEST
Knowledge ingestion
Index websites, selected URLs, internal documents, and other organizational sources within reusable knowledge collections.
ORGANIZE
Reusable collections
Group approved sources into governed collections that can support one or several assistants.
CONFIGURE
Assistant configuration
Define each assistant’s name, purpose, language behavior, prompts, welcome messages, skills, and access level.
RETRIEVE
Advanced retrieval
Reformulate queries, identify relevant sources, rank evidence, and retrieve the context needed for accurate responses.
GENERATE
Controlled generation
Generate answers, summaries, and content from selected organizational knowledge rather than unrestricted external information.
GOVERN
Central governance
Manage users, groups, permissions, collections, assistants, usage limits, audit trails, and feedback from one administration layer.
Modular architecture
Separate interfaces, governance, retrieval, and generation into manageable components
SAVIA is designed as a modular system in which each layer performs a defined role. This allows organizations to update models, data sources, interfaces, or processing services without rebuilding the entire platform.
01
User interface
Users access assistants through web applications, embedded interfaces, or connected enterprise systems.
02
SAVIA API
The central API manages authentication, access control, usage limits, assistant selection, and request routing.
03
Specialized services
Processing modules extract content, reformulate queries, retrieve evidence, evaluate relevance, and prepare context.
04
Controlled response
The system generates or returns a response according to the assistant’s knowledge, prompt, permissions, and assigned skills.
Specialized assistant capabilities
Configure different assistants for different users, departments, and workflows
SAVIA is designed to support assistants with distinct knowledge collections, system instructions, language behavior, access levels, and operational skills.
Q&A
Question answering
Respond to employee, customer, or citizen questions using approved documents and organizational sources.
SEARCH
Advanced search
Retrieve relevant documents and passages through natural-language queries rather than rigid keyword searches.
SUMMARY
Automatic summaries
Produce concise summaries of documents, source collections, search results, or operational information.
GENERATE
Content generation
Draft controlled content based on trusted organizational information, terminology, and predefined instructions.
LANGUAGE
Multilingual interaction
Configure prompts, interfaces, knowledge collections, and responses for multilingual organizations and users.
ROLES
Role-specific assistants
Create assistants for human resources, customer service, technical documentation, sales, compliance, public information, or internal support.
Security and access control
Enterprise assistants require governance at the user, assistant, collection, and interaction levels
SAVIA is designed around differentiated user profiles and granular permissions. Administrators can determine which groups may access a particular assistant, collection, or capability.
The platform also records queries, responses, feedback, and usage metrics, giving organizations a basis for auditing interactions and improving response quality.
This creates a more controlled alternative to sending internal knowledge directly into public generative AI interfaces with limited organizational oversight.
PROFILE
Differentiated roles
Support users, administrators, and platform-level administrators with separate responsibilities.
GROUP
Group permissions
Assign access to collections, assistants, and skills according to organizational groups.
TRACE
Interaction logging
Record incoming queries, generated responses, user feedback, and consumption data.
CONTROL
Deployment control
Connect SAVIA to private infrastructure, controlled models, and enterprise systems where required.
Knowledge management
Turn scattered enterprise content into reusable and maintainable knowledge collections
SAVIA treats organizational knowledge as a managed asset. Sources can be collected, grouped, updated, assigned to assistants, and reused across several business functions.
WEB
Website indexing
Index a complete website recursively or select individual pages and sources.
FILES
Internal documents
Connect approved files and document repositories to specialized assistants.
COLLECT
Knowledge collections
Group related sources into reusable collections aligned with departments or use cases.
UPDATE
Continuous maintenance
Review, replace, and update knowledge sources as organizational information changes.
Organizational applications
A common platform for multiple enterprise and public-sector assistants
The same governance and knowledge architecture can support several assistants, each with different users, instructions, collections, and permissions.
HR
Employee support
Answer questions about internal policies, procedures, benefits, onboarding, and organizational documentation.
SERVICE
Customer service
Provide grounded responses from product documentation, support materials, and approved service information.
PUBLIC
Citizen information
Help public administrations provide accessible answers based on official information and service documentation.
TECH
Technical documentation
Search manuals, specifications, knowledge bases, procedures, and engineering repositories.
SALES
Commercial support
Help sales teams retrieve approved product information, case studies, terminology, and proposal content.
COMPLIANCE
Policy and compliance
Assist users in locating internal policies, regulatory information, and controlled procedural guidance.
Pangeanic’s contribution
Combining multilingual data engineering, retrieval, generative AI, and controlled deployment
SAVIA builds on Pangeanic’s experience in language technology, AI Data Operations, multilingual knowledge processing, private infrastructure, and enterprise software development.
LEAD
Project coordination
Pangeanic coordinates the project, platform architecture, software development, and transition toward operational use.
NLP
Language processing
The platform applies multilingual retrieval, query reformulation, content processing, and response generation.
RAG
Retrieval architecture
SAVIA connects generative models to selected organizational knowledge through controlled retrieval workflows.
SECURE
Access and governance
The system includes authentication, groups, roles, permissions, usage controls, and traceability.
UX
Administration and chat
Web interfaces support assistant creation, knowledge management, dashboards, conversations, and feedback.
TRANSFER
Technology transfer
The project extends Pangeanic’s sovereign AI platform with reusable enterprise assistant and knowledge-management capabilities.
From research to deployment
Enterprise AI is moving toward smaller, specialized assistants connected to governed data
Organizations increasingly need systems built around a defined task, audience, knowledge base, and security model. A generic assistant cannot automatically reproduce the terminology, permissions, evidence standards, or operational context of every enterprise.
SAVIA provides a practical architecture for deploying several specialized assistants within the same governed environment. Each assistant can use different knowledge, instructions, permissions, and capabilities while sharing a common administrative layer.
This approach aligns with the broader shift toward task-specific models and systems that organizations can adapt, evaluate, and control.
Multilingual RAG Ground generative AI in trusted multilingual knowledge Sovereign AI systems Operate assistants within controlled infrastructure Datasets for AI Build and evaluate systems with governed domain data Model alignment Evaluate responses with human and domain expertise Anonymization Protect sensitive information before AI processing Enterprise AI consultation Define assistants, knowledge sources, and deployment requirements
Public project evidence
Funding, development scope, and published project information
The following public resources document SAVIA’s funding, development period, architecture, capabilities, and organizational objectives.
Official Pangeanic project article
SAVIA: AI Virtual Assistant Platform
Read the published project overview covering funding, architecture, knowledge management, permissions, and assistant capabilities.
Read the project article → Research portfolioResearch and Innovation Projects
Place SAVIA within Pangeanic’s broader portfolio of European, Spanish, and regional AI research projects.
Explore the project portfolio → Research activityResearch and Publications
Explore Pangeanic’s work in language technology, multilingual AI, privacy, data, and model evaluation.
Explore Pangeanic research →Continue exploring
From enterprise knowledge assistants to private and sovereign AI
Explore the project portfolio, private AI infrastructure, multilingual data, and model evaluation capabilities connected to SAVIA.
Research portfolio
Research and Innovation Projects
Explore the European, Spanish, and regional projects behind Pangeanic’s multilingual AI capabilities.
View all projects → Spanish industrial R&DPrivacy-Controlled AI Translation
Discover the CDTI project combining retrieval, terminology, quality estimation, and private adaptive translation.
Explore the CDTI project → Controlled infrastructureSovereign AI Systems
Operate assistants, models, knowledge bases, and data workflows within infrastructure controlled by your organization.
Explore sovereign AI → Grounded enterprise AIMultilingual RAG
Connect language models to governed multilingual knowledge for more accurate and traceable enterprise responses.
Explore multilingual RAG →Enterprise knowledge and private AI
Build specialized assistants around the knowledge, permissions, and infrastructure your organization controls
Pangeanic helps enterprises and public institutions define knowledge sources, retrieval systems, multilingual assistants, evaluation workflows, and private deployment architectures.