AI Agent Architecture & Infrastructure Consultant

Company Description DIREK LTD transforms buildings into interactive, data-driven collaborators that teams can engage with in natural language. The company connects to more than 80% of building systems on the market, including existing sensors, BMS, EMS and HVAC, then manages the full journey from live monitoring to deep analytics and expert reporting through its AI and Agentic AI stack. DIREK’s FM AI copilot provides facility, space and energy teams with clear, actionable insights instead of complex dashboards and spreadsheets. The organisation focuses on revealing what is really happening inside buildings, closing performance gaps, and enabling practical, affordable energy and space optimisation for a wide range of clients.

At DIREK, we are pushing the boundaries of spatial intelligence, IoT integration, and building management systems. We rely on a modern, robust tech stack to handle complex data streams and backend services. We are looking for an expert consultant to help us design and implement scalable AI agent architectures that integrate seamlessly into our existing infrastructure.

The Role

We are seeking an experienced AI Consultant to design multi-agent systems and the surrounding infrastructure (harnesses) required to evaluate, test, and deploy them reliably. You will not just be writing prompts; you will be defining the cognitive architecture, state machines, and memory systems for our agents, while ensuring they operate safely within our Google Cloud Platform (GCP) environment.

You will guide our technical team in establishing best practices for agent evaluation, deterministic testing of non-deterministic models, and observability.

Core Responsibilities

• Architectural Design: Design robust, scalable single and multi-agent architectures tailored to our specific use cases (e.g., automated reasoning over complex datasets).
• Harness Development: Build the testing and evaluation harnesses necessary to measure agent performance, track regressions, and ensure safe execution boundaries.
• Infrastructure Integration: Work alongside our team to integrate these agent workflows into our existing Python/FastAPI backend and GCP cloud architecture (e.g., Cloud Run, API Gateway).
• API & Tool Orchestration: Design agentic workflows that can reliably translate natural language questions into precise function calls against our FastAPI backend and time-series databases (e.g., TimescaleDB) to retrieve and synthesize answers.
• Data Ingestion & Code Interpretation: Architect secure, sandboxed agent environments capable of receiving unstructured or semi-structured user data (like .csv dumps), validating it, and writing/executing the necessary scripts (e.g., using Polars or Pandas) to format and integrate it into our analytical pipelines.
• Multi-Step Planning: Implement robust "Plan-and-Execute" architectures where the agent can break down a complex user request ("upload this energy usage CSV and compare it to last month's sensor data") into secure, sequential backend operations.
• Mentorship & Strategy: Provide actionable guidance on the "build vs. buy" decisions for agent tooling and establish CI/CD pipelines for LLM-driven applications.

Required Skills & Qualifications

1. Agent Architectures & Orchestration

• Frameworks & Routing: Deep, practical experience with modern agent orchestration frameworks (e.g., LangGraph, AutoGen) and the ability to build custom semantic routers to direct user queries to the appropriate backend service.
• API & Tool Calling: Proven ability to architect workflows that reliably translate natural language into precise function calls, with strict enforcement of JSON outputs that adhere to complex OpenAPI schemas.
• State Management & Recovery: Expertise in designing state-machine-driven flows with self-correction loops—enabling agents to autonomously catch API errors or script failures and rewrite their requests without user intervention.

2. Secure Data Ingestion & Code Execution

• Code Interpreter Patterns: Experience building and deploying environments where agents can safely write and execute Python code to ingest, parse, and clean unstructured or semi-structured data (e.g., user-uploaded .csv files).
• Sandboxing & Security: Strong understanding of isolated execution environments (e.g., secure Docker containers, WebAssembly) to guarantee that agent-generated scripts cannot compromise core cloud infrastructure or access unauthorized environment variables.
• Data Manipulation & Validation: High proficiency with high-performance data libraries (e.g., Polars, Pandas) and validation frameworks (e.g., Pydantic, Great Expectations) to ensure data quality before it enters our backend analytical pipelines.

3. Harnesses, Evaluation & Observability (EvalOps)

• Deterministic Testing: Experience building robust evaluation harnesses to measure agent performance and track regressions using frameworks like Ragas, Promptfoo, or DeepEval.
• Tracing: Implementation of comprehensive observability and tracing for complex, multi-step agent trajectories using tools like LangSmith, Phoenix, or OpenTelemetry.

4. Engineering & Backend Integration

• Core Languages: Expert-level Python.
• Cloud & Infrastructure: Strong track record of deploying containerized microservices within Google Cloud Platform (GCP) (e.g., Cloud Run, API Gateway).
• Backend Systems: Familiarity with modern API development (FastAPI) and integrating with time-series databases (e.g., TimescaleDB) and event-driven architectures (e.g., Kafka) for high-throughput data streams.