Job Title: Agentic AI Architect Location: Preference is Charlotte, NC but open to remote Duration:6+Months
This role requires deep expertise in Agentic AI, MCP, and related emerging technologies.
Build end-to-end architecture for an enterprise agentic AI platform — multi-agent orchestration, tool/MCP ecosystems, RAG, evaluation, and production guardrails. Define the reference architecture and lead technical decisions across squads. Key role
Design multi-agent systems (planner/executor/critic, supervisor-worker, swarm patterns) using frameworks such as Claude Agent SDK, Lang Graph
Define the MCP (Model Context Protocol) tool catalog, schema, server topology, authn/authz model, and governance for tool exposure across agents.
Architect hybrid RAG (vector + lexical/BM25, e.g., pgvector + Postgres) with chunking, re-ranking, citation, and freshness strategies.
Establish patterns for memory (short/long-term, episodic), state persistence, human-in-the-loop gates, and deterministic checkpoints.
Define model routing (Opus/Sonnet/Haiku), cost/latency budgets, and fallback strategies.
Set evaluation strategy: golden sets, LLM-as-judge, trajectory eval, regression suites, and observability (LangSmith/Lang Fuse/CloudWatch).
Drive security/compliance: prompt-injection defenses, secret hygiene, PII handling, data residency, model risk management.
Author ADRs, HLDs, and reference implementations; mentor developers and review code.
Hands-on with at least two agentic frameworks (Claude Agent SDK, Lang Graph, AutoGen).
RAG at scale: pgvector / OpenSearch / Pinecone, embedding model selection, hybrid retrieval, re-ranking.
MCP server/client design and tool integration patterns.
Applicant Notices & Disclaimers
For information on benefits, equal opportunity employment, and location-specific applicant notices, click here
At SPECTRAFORCE, we are committed to maintaining a workplace that ensures fair compensation and wage transparency in adherence with all applicable state and local laws.This position's pay range is $96.00/hr - $100.00/hr.