| Description: |
AI Architect I'm currently working with a leading consulting client as they are seeking an AI Architect for a project related collaboration. Initial contract length: 12 months Working Model: Hybrid (2-3 days per week) Rate: 550-600 About the Role - Hands-on senior IC role designing, building, and productionising enterprise-grade multi-agent AI systems at scale
- Operates across the full agentic stack - orchestration, model management, RAG, observability, and evaluation
Key Responsibilities - Architect production-grade multi-agent workflows using frameworks such as LangGraph, CrewAI, AutoGen etc.
- Design and implement agent orchestration patterns - intent detection, dynamic agent routing, conversation state management, and hierarchical multi-agent topologies
- Build and govern MCP (Model Context Protocol) Servers exposing enterprise tools, internal systems, APIs, and data sources as structured, auditable agent tool surfaces. Knowledge of A2A is an advantage.
- Develop against and configure AI Gateway platforms (LiteLLM, AWS Bedrock, Azure API Management, Apigee AI Extensions) for multi-model routing, cost controls, quota enforcement, and model fallback
- Integrate Responsible AI guardrails - PII masking, content moderation, toxicity filtering, and policy enforcement - across the model invocation layer
- Own RAG pipeline design and optimisation - chunking strategy, embedding model selection, hybrid retrieval, reranking, hallucination mitigation etc.
- Build and maintain LLM evaluation frameworks - golden test sets, regression suites, LLM-as-a-judge, online A/B and shadow evals, red-teaming for prompt injection, jailbreaks, and adversarial inputs
- Own and build LLM observability using platforms such as Langfuse, etc.
- Contribute to CI/CD pipelines for agent systems - containerised deployments, automated behavioural testing, feature flag management, and progressive rollout patterns
- Apply security-first engineering aligned with Agent Development Lifecycle (ADLC) principles - agent identity, least-privilege tool access, sandboxed execution, and full audit traceability etc.
 |
