Aquisição de serviços de conceção, desenvolvimento, implementação e disponibilização online de um assistente virtual para o pc 360º

The contract timeline is as follows:

Contract start date:

The contract begins either from the award date or with its signing.

Contract end date:

The execution of the contract cannot extend beyond December 31, 2026.

Total duration:

There are conflicting statements regarding the total duration. One source indicates a period of four (4) months starting from the award date, while others state a maximum execution period of 2 months.

Key milestones/phases:

The project is structured into eight minimum phases, some of which may run in parallel:

1. Requirements gathering and validation: This phase is scheduled for Month 1, Weeks 1-2, involving a detailed review of requirements, confirmation of access, and validation of external sources.

2. Architecture: This phase, set for Month 1, Weeks 2-4, focuses on defining and validating the complete technical architecture of the solution.

3. MVP Implementation: Scheduled for Month 2, Weeks 1-4, this phase involves developing the minimum viable product with core functionalities.

4. Integration and indexing: This phase takes place from Month 2-3, Weeks 3-10, running in parallel with Phase 3 from Week 3 of Month 2, and includes integration with the archive, digital collection ingestion and indexing pipeline, vector database configuration, and Azure Computer Vision integration.

5. Testing: This phase, from Month 3, Weeks 2-4, covers technical and functional tests, including performance, access control, GDPR, and embargo rules.

6. Restricted pilot: Occurring from Month 3-4, Weeks 4-14, this involves making the solution available to a restricted group of users for functional validation and feedback, followed by anomaly correction.

7. Final version: This phase, scheduled for Month 4, Weeks 2-3, involves complete implementation in a production environment with all mandatory functionalities validated and acceptance criteria met.

8. Training and transition: From Month 4, Weeks 3-4, this includes training the contracting entity's technical team and full knowledge transfer, along with the delivery of complete technical documentation and operation scripts.

The core scope of work for this tender involves the conception, development, implementation, and online availability of a virtual assistant for the PC 360º. This is classified as an acquisition of services, specifically application software programming services (CPV 72212000). The execution period for the contract is four months from the date of award.

The project is structured into several minimum phases:

• Requirement Survey and Validation (Month 1, Weeks 1-2): Detailed review of requirements, confirmation of access, and validation of external sources.
• Architecture Definition (Month 1, Weeks 2-4): Defining and validating the complete technical architecture, including the technology stack, data model, RAG pipeline, and infrastructure.
• MVP (Minimum Viable Product) Implementation (Month 2, Weeks 1-4): Development of core functionalities such as RAG chat, text search, authentication, access control, and a basic administration panel.
• Integration and Indexing (Months 2-3, Weeks 3-10): Integration with "arquiv@", development of ingestion and indexing pipelines for digital assets, vector database configuration, and integration with Azure Computer Vision for image pre-processing.
• Testing (Month 3, Weeks 2-4): Conducting technical and functional tests, including performance (response time <5s), access control, GDPR compliance, and embargo rules.
• Restricted Pilot (Months 3-4, Weeks 4-14): Deployment to a limited user group for functional validation, feedback collection, and anomaly correction.
• Final Version Implementation (Month 4, Weeks 2-3): Full implementation in a production environment with all mandatory functionalities validated.
• Training and Transition (Month 4, Weeks 3-4): Training the contracting entity's technical team and comprehensive knowledge transfer, including technical documentation and operation scripts.

Core Deliverables:

• Complete source code and components necessary for local execution, including installation and configuration scripts, ensuring the core does not depend on an external cloud.
• Ingestion/indexing pipeline for text and images, a vector database, and mechanisms for updates.
• A web interface and an administrative panel.
• Detailed technical documentation covering installation, operation, maintenance, security, and updates.
• A multimodal functional demonstration (text + images) before final acceptance.

Key functionalities and requirements explicitly mentioned include:

• Local execution for indexing, embeddings, and search, with only RAG context excerpts (not full documents or personal data) being sent to the LLM cloud.
• Textual and image search capabilities.
• RAG with citation.
• OCR functionality for scanned documents.
• Handling of restricted content.
• Features for favorites, sharing, and an admin panel for managing external source allowlists.
• Logs/auditing, transcription of scanned documents, and performance metrics (response time under 5 seconds for typical queries).
• Visual timeline/map exploration and multilingual support beyond Portuguese (PT-PT) and English (EN-UK).
• Prohibition of sending complete documents, personal data, or sensitive metadata to the LLM cloud, allowing only RAG context excerpts.
• Text generation based on properties and on-demand deepening.
• Export functionalities (PDF/CSV/JSON) and reporting.
• Metrics and monitoring (KPIs, logs, alerts).