Context

As part of the evolution of AI projects within Forem, the Information Systems Department wishes to enhance its service offering by integrating a vector and NoSQL database solution. The chosen solution is Azure Cosmos DB.

The anticipated increase in workload (data volume, queries, new AI use cases) requires ensuring a scalable, high-performance, and optimized solution for large-scale use cases (AI, RAG, hybrid search, etc.).

Mission Objective

Assess, define, and formulate operational recommendations to industrialize the use of Cosmos DB as a vector database in an AI context, ensuring:

• Performance and robustness
• Cost control
• Long-term maintainability

Scope of the Mission

The consultant will work on the following areas:

• Analysis
• of the current architecture and planned uses.
• Identification of constraints and risks related to:
• Scaling up (data volume, QPS, latency)
• Management of vector indexes
• Data distribution
• Cosmos DB-specific limitations
• Security risks
• Recommendations on:
• Choice of architecture (modeling, partitioning, indexing, RAG, hybrid search)
• Best practices for industrialization
• Scalability and resilience strategies
• Cost control
• Clarification of the roles and responsibilities of the ISD management (infra, data, development) around the use and governance of Cosmos DB and associated technologies
• Proposal for a monitoring and observability approach (performance, consumption, alerting)
• Identification of operational points of attention (operations, maintenance, scalability)
• Concise and actionable report for the ISD

Expected Deliverables

• Document with clear and prioritized recommendations
• Target architecture diagram to facilitate implementation
• Concrete recommendations, directly usable by technical teams
• Oral presentation to stakeholders

Execution Details

Duration: Maximum 10 days, spread between mid-January and the end of February, depending on stakeholder constraints.

On-site presence required for:

• Initial scoping
• Meetings for reporting and presentation of recommendations

The remainder of the mission may be carried out remotely.

The consultant may rely on internal teams (data scientists, architects, technical experts, etc.) to successfully complete the mission.