Solution Architect - Agent Catalogue
Provide solution architecture and technical design assurance for agentic artificial intelligence solutions across Unilever.
- Bangalore, India
- Posted Nov 11, 2026
Responsibilities
- Provide solution architecture for artificial intelligence and agentic use cases across multiple technologies and platforms.
- Translate business requirements into robust technical architectures covering agents, models, skills and tools, data, integration and orchestration.
- Conduct technical design reviews and ensure solutions meet agreed architecture and engineering standards before build and production.
- Determine when requirements should be met through native platform capabilities, reusable enterprise capabilities or custom pro-code development, keeping architectures portable and avoiding unnecessary platform lock-in.
- Define and apply architecture patterns covering orchestration, grounding, agent-to-agent interaction, human-in-the-loop, application programming interfaces, Model Context Protocol (MCP) and tool integration, and deterministic automation.
- Make sure agentic solutions are designed for enterprise scale rather than individual proofs of concept or isolated use cases.
- Assess designs for scalability, resilience, observability, maintainability, security and cost.
- Monitor the evolution of models, agent frameworks and platform capabilities, and determine where existing architecture patterns need to evolve.
- Partner with the Senior Manager, AI Solution Architecture and Enterprise Architecture to convert emerging approaches into approved reusable patterns.
- Support the progression of solutions towards greater autonomy while maintaining appropriate control, audit and human intervention points.
- Embed security, privacy, Responsible AI and governance requirements directly into solution design.
- Make sure agent identity, permissions, data access, tool access and execution boundaries are appropriately designed.
- Maintain traceability of architecture decisions and exceptions, identify architectural risks early and drive remediation before production deployment.
- Make sure reusable assets have clear technical ownership and lifecycle accountability.
- Anchor token optimisation across agentic workloads, analysing excessive context, prompts, agent interactions and model calls and recommending architectural optimisation.
- Drive patterns such as context reduction, better retrieval, reusable system instructions, efficient agent and tool interaction and right-sized model selection.
- Work with solution teams to improve cost per outcome, not simply model cost per token, and identify where smaller, lower-cost models can safely replace larger models for defined task classes.
- Build innovative solutions that improve efficacy, and feed optimisation learnings back into architecture standards so improvements become reusable patterns rather than one-off fixes.
- Drive ongoing hydration of the enterprise Agent Catalogue and Skill Registry, making sure reusable agents, skills, tools, prompts, integrations, architecture patterns and accelerators emerging from delivery are identified and catalogued.
- Establish metadata that makes assets discoverable and reusable, including capability, owner, technology, version, dependencies, applicability and lifecycle status.
- Assess new demand against existing assets before a new build is authorised.
- Identify duplicate or overlapping agents and skills across platforms and domains, and recommend consolidation where multiple teams have developed similar capabilities.
- Work with business and technical subject matter experts to keep catalogue content current and usable.
- Convert recurring solution requirements into reusable enterprise architecture and engineering patterns.
- Encourage composition of solutions from approved skills and reusable building blocks rather than monolithic agent implementations.
- Partner with engineering teams to industrialise successful proofs of value into repeatable accelerators.
- Track adoption of reusable components, identify barriers to reuse and enable a build once, reuse across products and platforms approach wherever technically appropriate.
Requirements
Nice to have: Microsoft Copilot Studio, Model Context Protocol servers, plugins, connectors, enterprise tool integration, LLMOps, generative artificial intelligence operations, AI observability, model evaluation, benchmarking, prompt engineering, AI financial operations (FinOps), cost-management, knowledge graphs, semantic search, DevOps, continuous integration, continuous delivery, autonomous architectures, NoOps, developer portals, internal technology marketplaces, fast-moving consumer goods, retail, eCommerce, global manufacturing organisations
Skills
Full description
Sign up for a free job alert to see Unilever's profile, website and LinkedIn, and the same for every employer on Referso Jobs.