About this role
XenonStack seeks a Solution Architect, Agentic Systems to design and lead enterprise-scale multi-agent architectures. The role combines LLMs, tools, memory and orchestration frameworks to create reliable, adaptive and compliant AI agents. The position is listed in Mohali, India, as full-time, with a salary of 8–15 LPA.
Responsibilities
- Define scalable, modular blueprints for agent workflows, reasoning loops, tool integration, memory and multi-agent orchestration.
- Design context pipelines using RAG, knowledge graphs, APIs and short-term, episodic and long-term memory; manage tokens and context allocation for reliability and cost efficiency.
- Connect LLMs with enterprise data sources and observability layers, working with AgentOps engineers on deployment and monitoring.
- Implement safety, compliance and brand-alignment guardrails; ensure designs meet Responsible AI and governance standards.
- Evaluate emerging approaches including LangGraph, MCP, AgentBridge and A2A messaging. Partner with product managers and executives on enterprise workflows, and mentor engineers and product teams on accuracy, cost, latency and compliance trade-offs.
Required qualifications
The qualifications section specifies 6–10 years in AI/ML engineering, systems architecture or enterprise software design, although the job-information field lists work experience as 4–5 years. Candidates need knowledge of LLM architectures, LangChain, LangGraph, LlamaIndex, agent design patterns, context engineering, RAG, vector and knowledge databases, and multi-agent orchestration frameworks including MCP, A2A messaging and AgentBridge. Python proficiency, familiarity with AWS, GCP, Azure, Kubernetes and Docker, and understanding of Responsible AI and compliance-driven design are also required.
Preferred qualifications
Reinforcement learning experience, including RLHF, RLAIF or reward modeling; enterprise deployments in BFSI, GRC, SOC or FinOps; prior Solutions Architect, AI Architect or Technical Lead experience; and open-source contributions to multi-agent or LLM-orchestration frameworks.