AI Solution Architect - Agentic AI

iSystems Ltd; Karachi Division; On-site; PKR 3,500,000 - 6,500,000; Full time

Job summary

iSystems Ltd is seeking an experienced Architect to lead the design of agentic, multi-agent systems and the orchestration layer. You will specify agent roles, coordination patterns, memory strategies, and tool-use permissions for reliable, safe autonomous systems.

You will mentor engineers, define guardrails, and present risk/benefit tradeoffs to clients. Strong architecture governance and ADR documentation are essential.

Qualifications

• 10–15+ yrs software/AI architecture experience.
• 2+ yrs specifically in agentic/multi-agent systems.
• Deep knowledge of agent orchestration frameworks and failure modes.
• Experience with managed agentic platforms (Azure/AWS/Google).
• Strong systems-design fundamentals: distributed systems, state machines, event-driven architecture.
• Experience memory architectures for long-running agents.
• Familiarity with agent evaluation methodologies and benchmarking.
• Working knowledge of guardrail/governance tooling for agentic systems.
• Ability to defend architecture decisions under adversarial challenge.
• Experience mentoring/leading a technical team.

Responsibilities

• Design multi-agent architectures, agent roles, coordination patterns, handoff logic.
• Own agentic workflow design: task decomposition, tool-calling, memory/state management including knowledge graphs for structured reasoning.
• Define guardrails: tool-use permission, kill switches, escalation paths for agent failure.
• Architect evaluation framework for agentic systems in partnership with AI Assurance/QA.
• Lead technical design reviews for agent-based solutions before build.
• Design fallback and human-in-the-loop patterns for high-stakes agent decisions.
• Mentor GenAI Engineers on architecture patterns and review their designs.
• Present agentic system risk/benefit tradeoffs to client stakeholders.
• Own architecture governance for agentic builds across the practice.
• Facilitate cross-team design reviews involving Security, Governance, Platform Engineering.

Skills

Multi-agent architectures

Agent orchestration

Distributed systems

Memory architectures

ADR/documentation

Tools

Azure AI Foundry Agent Service

AWS Bedrock Agents/Strands Agents

Google Vertex AI Agent Builder

LangGraph

AutoGen

CrewAI

Job description

Architects’ multi-agent and agentic workflow systems — the orchestration layer, memory design, and tool-use permission that keep autonomous systems reliable and safe.

KEY RESPONSIBILITIES

• Design multi-agent architectures — agent roles, coordination patterns, handoff logic (LangGraph, AutoGen, CrewAI)
• Own agentic workflow design: task decomposition, tool-calling, memory/state management, including knowledge graphs for structured reasoning and long-term memory
• Define guardrails: tool-use permission, kill switches, escalation paths for agent failure
• Architect the evaluation framework for agentic systems in partnership with AI Assurance/QA
• Lead technical design reviews for agent-based solutions before they reach build
• Design fallback and human-in-the-loop patterns for high-stakes agent decisions
• Mentor GenAI Engineers on architecture patterns and review their designs
• Present agentic system risk/benefit tradeoffs to client technical and business stakeholders
• Own architecture governance for agentic builds across the practice
• Facilitate cross-team design reviews involving Security, Governance, and Platform Engineering

REQUIREMENTS & SKILLS

• 10–15+ yrs software/AI architecture, with 2+ yrs specifically in agentic/multi-agent systems
• Deep knowledge of agent orchestration frameworks and their failure modes (loops, deadlocks, drift)
• Evaluates and architects on managed agentic platforms as per area of specialization including Microsoft Azure AI Foundry Agent Service, AWS Bedrock Agents/Strands Agents, and Google Vertex AI Agent Builder; open-source orchestration (LangGraph, AutoGen, CrewAI) etc.
• Strong systems-design fundamentals — distributed systems, state machines, event-driven architecture
• Experience designing memory architectures (vector + graph-based) for long-running agents
• Familiarity with agent evaluation methodologies and benchmarking approaches
• Working knowledge of guardrail/governance tooling for agentic systems
• Can defend an architecture decision under adversarial technical challenge from a client's own architects
• Writes clear ADRs and architecture documentation that both engineers and executives can follow
• Experience mentoring/leading a technical team, not just designing solo
• Comfortable being the accountable technical owner when an agentic system fails in production