Policy Advisor Tool for CSB

The CSB Policy Advisor initiative aims to:

• Develop a searchable, centralized repository of Arabic jurisprudence documents and policies.
• Use advanced AI technologies, including a fine-tuned Large Language Model (LLM), to process and analyze Arabic texts.
• Create a web-based interface for querying and obtaining detailed, actionable insights from various jurisprudence documents.
• Facilitate efficient and transparent decision-making in the public sector.

1. Scope of Work

System Development

• CSB Jurisprudence Library
  • Centralized repository of Arabic jurisprudence documents.
  • Indexing and categorization tailored for Arabic-language complexities.
  • Seamless integration of periodic content updates.
• AI-Driven Insights Using Customized LLM
  • Leveraging frameworks such as OpenAI’s GPT, Falcon, or LLaMA models, fine-tuned for Jurisprudence terminologies and syntax in Arabic.
• Web-Based Prompting System
  • Natural language query supports Arabic for retrieving insights.
  • Context-aware prompting to refine and optimize responses based on user input.
  • Multilingual capabilities for accommodating bilingual legal research needs (Arabic and English).

Document Feeding Interface

• Arabic Document Ingestion System
  • Metadata tagging for Arabic content, ensuring improved categorization and searchability.
  • API-enabled integration for seamless input from external Jurisprudence repositories.
  • Batch upload functionality to manage bulk documents ingestion efficiently. This functionality should be developed as a secure web interface allowing CSB authorized personnel to upload as many as needed and in 1 batch files into the repo.

Technical Architecture

• Frontend:
  • Responsive web interface in Arabic and English using modern frameworks to be integrated, into the existing CSB Portal , by the provider with the guidance of CSB Team.
  • Real-time query response with AI-generated insights and visualizations.
• Backend:
  • Scalable cloud-based architecture optimized for processing Arabic texts.
  • Use of Elasticsearch and similar tools adapted for Arabic indexing and retrieval.
  • Integration of AI/ML frameworks such as TensorFlow or PyTorch for model deployment.
• Language Model (LLM) Specifications:
  • Pre-trained on diverse Arabic legal corpora, fine-tuned for domain specificity.
  • Continuous learning capabilities to incorporate new documents and legal updates.
  • Secure hosting environment ensuring data integrity and compliance with local regulations.
• Data Storage and Security:
  • Hybrid storage model with relational and non-relational databases for structured and unstructured Arabic data.
  • End-to-end encryption of sensitive data during storage and transit.
  • Compliance with data protection and privacy standards in Lebanon.

2. Deliverables

• Fully functional CSB Jurisprudence Library tailored for Arabic content.
• An AI-powered digital policy assistant with robust Arabic query capabilities.
• Web-based interface for interactive querying and insights in Arabic and English.
• Web based automated Arabic documents ingestion and processing system.
• Comprehensive user and technical documentation in English.
• Strictly onsite training and possible hybrid technical support for CSB staff.
• The solution delivered should encompass 3 years free of charge technical support for the solution delivered features, a proper SLA to govern this requirement should be presented within the technical proposal.

3. Duration and Schedule

The project timeline is 3-5 months:

• Month 1-2: System design, requirements gathering, and initial Arabic data preprocessing.
• Month 2-3: Development of AI models, Arabic text indexing, and web interface design.
• Month 3-5: System integration, testing, user training, and deployment.

4. Methodology

The approach will include:

• Close collaboration with the EF component manager, IT expert, Website developer as well as CSB IT staff for technical clarifications.
• Iterative development and early previews to ensure alignment.
• Final validation by the EF Team leader and IT expert before submission.

5. General Requirements

• Proposed AI Models must have at least 2 years of end of life (EOL)
• Solutions requirements are highly preferred to be open source, alternatively any proposed license should be quoted per year and show a 3-year pricing projection within the financial proposal.
• Solution hosting will be at OGERO (The republic of Lebanon official hosting provider), the provider shall detail the needed sever specs to host the solution. Alternatively, cloud-based hosting specs can be provided as a possible route yet not binding to CSB or Expertise France thus the OGERO hosting should always be considered when designing and deploying the solution.
• The proposed solution must include 3 years’ free maintenance and support with a proper SLA.
• Delivery is onsite at Central Service Board (CSB) premise located in Beirut - Verdun– Lebanon (location: https://maps.app.goo.gl/wDzfvf3SAUTMFoAc9 ).
• None of the overall solution components can be installed or delivered remotely. All must be installed in presence and collaboration with CSB IT Staff.
• CSB working hours are Monday to Friday from 800 AM – 3:00 PM excluding public holidays as depicted by the Republic of Lebanon, all deployment and training works must be conducted within this timeframe.

6. Evaluation

The best value for money is established by weighing technical quality against price on a 60/40 basis. The quality of each technical and financial offer will be evaluated in accordance with the following award criteria and the weighting:

CRITERIA

WEIGHTS

Requirements and experience

60

Service provider Experience & References

(15)

Technical and Methodology Capabilities

(30)

Proposed delivery timeline

(15)

Price (including TCO)

40

Tenders will be appraised and given a score of up to 100 points according to these criteria.

NB:

• Only tenders with scores of at least 45 points on technical evaluation qualify for the financial evaluation.
• No other award criteria will be used. The award criteria will be examined in accordance with the requirements indicated in the Terms of Reference.

7. Qualifications and Experience

• The service provider must be a registered company with at least of 3-5 years of proven experience in delivering similar solutions.
• Demonstrated experience in developing comprehensive similar solutions particularly open source-based solutions.
• The service provider must provide evidence of at least 2 similar projects they have successfully completed. For each project, detailed descriptions of the work performed, the services provided, and the outcomes achieved should be submitted. This may include case studies, project reports, or client testimonials.

8. Place, duration, and terms of performance

Remote with onsite visits and follow-up at Civil Service Board (CSB) when needed. Training and handovers should be done solely onsite at CSB on a mutually agreed date/time between the provider and CSB staff. No Remote training or handover are acceptable.

2)            Start date: 25 June 2025

3)            Delivery date: 3-4 Months

9. Reporting and Supervision

The provider will work under the supervision of the Civil Service Board (CSB) Component Manager and closely with the project's designated IT expert and CSB staff.

10. Confidentiality and Ownership

The provider will submit updates every 14 days, including Arabic LLM training progress, document ingestion milestones, and system testing results. Regular meetings will ensure alignment with CSB goals and address stakeholders’ feedback.