Principal AI Architect

We operate at the intersection of data foundations and applied AI
, helping clients move from research concepts to real, deployed systems.

We are a small, sharp consulting team. We ship real systems, not just decks. As a Principal AI Architect
, you are the senior technical authority responsible for ensuring our AI solutions are not just impressive demos, but durable, governable, and operationally sound.

This is a hands-on principal IC role
, not a people-management position.

You are the technical decision-maker for AI delivery  bridge the gap between the art of the possible and production reality, owning decisions across model selection, RAG strategy, agent orchestration, security posture, and deployment architecture
.

You lead by doing. You write the core scaffolding code, define the architectural patterns others extend, and act as the final technical gatekeeper for production AI releases. When projects hit complexity cliffs, you are called in to unblock them.

While you do not formally manage people, you are expected to teach, mentor, and elevate the technical bar across the team.

• Architect agentic workflows: Design multi-step, stateful agent systems using Lang Graph, CrewAI, Semantic Kernel, or equivalent frameworks.
• Beyond prompting: Define how agents reason, plan, and act. Define tool selection, memory, guardrails, and failure handling.
• Application Logic & Cognitive Architecture: Build the complex chains, prompt orchestration, reasoning loops, and multi-agent workflows. Maintain conversation state, manage token limits, and handle tool-calling errors gracefully in production.
• Advanced RAG: Design retrieval systems end-to-end: chunking strategy, embeddings, indexing, hybrid search, grounding, and citation.
• Evaluation & Metrics Ownership: Design robust evaluation frameworks (LLM-as-a-Judge, Ragas, Tru Lens, Arize) to measure model performance, latency, and accuracy before production deployment. Own the 'truth' of the system and prove that AI solutions work reliably.
• Custom tooling: Build APIs and tools when off-the-shelf connectors fall short.

• Hands-on ownership: Personally implement high-impact components, core services, agent frameworks, complex integrations, and automation logic.
• Rapid Prototyping for Presales: Quickly build working demos or POC’s to validate feasibility and secure client engagements.
• Collaboration with Engineers:
• Work closely with Engineers on pipelines and infrastructure while taking ownership of cloud, agent orchestration, and deployment architecture.
• Infrastructure & deployment: Define and contribute to IaC (Terraform), CI/CD pipelines, environment strategy, and release processes.
• Cloud focus: Lead both GCP and Microsoft Azure technical work, filling gaps where our team has limited experience.
• Security & governance: Own the production-readiness checklist for AI systems, including PII handling, IAM, VPC boundaries, secrets management, and auditability.
• Engineering validation: Build rapid technical spikes and POCs to move initiatives from assumptions to proof and de-risk delivery.

You hold final technical sign-off on AI deployments.

• Pre-sales feasibility: Partner with the Director of Engineering and leadership to assess opportunities, prevent over-selling, and define realistic delivery paths.
• Reusable accelerators: Convert successful project patterns into reusable modules, templates, and reference architectures that accelerate future delivery.
• Client advisory: Translate technical trade-offs (latency vs. accuracy, cost vs. performance, managed vs. custom) into business-relevant decisions for senior stakeholders.
• Dedicate approximately 20% of your time to mentoring team members, performing code reviews, and shaping Fastloop’s AI standards, frameworks, and internal best practices.

• Production experience with Vertex AI, OpenAI, Azure AI
• Strong evaluation discipline using Lang Smith, Arize Phoenix, or equivalent
• Expert in Cognitive Architectures for multi-agent systems, using advanced patterns such as planning, reflection, and tool-use to solve…