Posted by Creative Chaos on 23 February 2026, 269 days ago. Still on their Workable board when we checked 45 min ago. Half of Creative Chaos's roles come down within 288 days.
Lead cross-functional teams to build AI-native products. Design and ship RAG/agent systems using various models and retrieval methods. Ensure safe AI operation with evals, guardrails, and observability. Own delivery outcomes and be client-facing. Requires 8-12+ years of engineering experience with 4+ years leading multi-team delivery.
Your experience, ready for this role
Answer a few questions about your real work. Review the additions and download your CV.
Job Summary:
As an AI Architect you will build AI-native products. You’ll lead cross-functional Innovation Delivery Squads—owning outcomes end-to-end across web, mobile, AI agents, and streaming backends. You’re a hands-on technical leader who can scope, architect, staff, and ship; then run the product safely at scale.
Job Responsibilities:
Stand up and run squads (Discovery → Prototype → Product → Platform & SRE).
Design and ship RAG/agent systems: pick models (e.g., Anthropic Claude, OpenAI, Google, or open-weights like Llama/Mistral), define tools/functions, and choose retrieval (default Postgres + pgvector, scale to Weaviate/Qdrant/Pinecone when needed).
Operate AI safely: evals & guardrails, structured outputs (JSON/Schema), PII redaction, refusal policies, cost/latency budgets, and LLM observability.
Own delivery outcomes: SLOs, quality, cost, velocity; release with feature flags and canaries.
Be client-facing: discovery, scoping, SoW, roadmap, QBRs.
Hire/coach Tech Leads, EMs, and PMs; level up practices.
Requirements
8–12+ yrs engineering; 4+ yrs leading multi-team delivery; shipped production web/mobile systems at scale.
Shipped at least one production AI app using Claude/GPT/Gemini/Llama/Mistral, backed by retrieval (pgvector or a vector DB) and a basic eval/guardrail pipeline.
Implemented orchestration (LangGraph/DSPy or Temporal for durable workflows), rerankers (e.g., Cohere/Jina/Voyage), and prompt/tool versioning.
Built with modern cloud + data: serverless/K8s, Terraform, OpenTelemetry, feature flags/experimentation.
Excellent client communication and commercial sense (SoWs, staffing, utilization).
Tech stack (you have hands on experience)
Models: Anthropic Claude; OpenAI; Google; open-weights (Llama, Mistral).
Orchestration & agents: LangGraph (or DSPy) for graphs; Temporal for durable, long-running tasks and SLAs.
Retrieval: Postgres + pgvector (default); Weaviate/Qdrant/Pinecone when scale/ops require; hybrid search with OpenSearch/Typesense.
Embeddings / rerankers: OpenAI/Voyage/E5/BGE; Cohere/Jina/Voyage rerank.
Guardrails & evals: JSON/Pydantic schemas, red-team sets, promptfoo/Ragas/DeepEval; content/PII filters.
Observability: OpenTelemetry traces incl. prompt/tool spans; Langfuse/Arize Phoenix (or equivalent) + Sentry/Grafana.
App & data: Next.js 15 (RSC), TypeScript/Go/Python; Postgres; Kafka/Redpanda/NATS; dbt/lakehouse optional.
Ops: Cloud Run/ECS/K8s; Terraform/OpenTofu; GitHub Actions; LaunchDarkly/Unleash; Statsig/GrowthBook.
Copied from Creative Chaos’s own board, not rewritten.
Also open at Creative Chaos
Why this page exists
We read companies’ own hiring systems every hour, 2,106 of them, and show a job the hour it opens instead of when a job board gets around to indexing it. We saw it 7 months after it went up.