Full Stack Python Engineer | NearShore | Remote - Build Circle

Build Circle is a tech consultancy founded by developers and consultants based in London.  We work with our clients to help bridge gaps in technology, accelerate delivery, improve code quality and enable the scaling of platforms and products. Our competitors include Equal Experts, 101 Ways and ThoughtWorks to name a few.

We have recently achieved advanced AWS partner status - you can find out a bit more about what we do here: https://www.buildcircle.co/what-we-do

Build circle is currently hiring for both permanent and contract consultants from remote locations across a variety of technical stacks and disciplines. We are a remote first company and are looking to engage with the best technical consultants and engineers on the global market to help us service our clients and who can provide hands-on technical expertise to the standard that Build Circle as a tech partner is known for. 

This role is for a Full Stack Python Engineer working with one of our clients in the media and content creation space. The organization is exploring how AI agents can be introduced to reduce operational costs and accelerate internal processes.

The current team has strong expertise in AI/LLMs and system architecture, and is primarily responsible for building the Java-based orchestration layer and managing the LLM provider and prompt ecosystem. They are now looking to add an engineer who can design and deliver production-grade Python agent services that integrate with this architecture.

Key Responsibilities
1. Build and operate production-grade agent services
Design, develop, and maintain containerized Python agent services built for production use — not prototypes. Ensure services include robust error handling, retry logic, graceful degradation, structured logging, health checks, and operational observability. Implement agents using a shared blueprint/library architecture to standardize authentication, messaging, and monitoring patterns across the system.

2. Own the Planning → Generation → Validation execution loop
Develop and maintain the core execution loop that powers the agent system. This includes enabling the Planning Agent to transform content analysis, governance rules, and user requirements into dynamic, mutable execution plans. Ensure the system can safely add, modify, or reorder tasks during execution based on real-time quality evaluation while maintaining clear exit conditions (e.g., human escalation, quality thresholds met, or system timeouts).

3. Design and maintain LangGraph orchestration workflows
Implement and manage LangGraph-based orchestration graphs that define how agents reason, invoke tools, and handle branching logic. Ensure workflows are modular, testable, and maintainable, with clear state transitions for scenarios such as sensitive data detection, policy enforcement, or content-specific processing paths.

4. Implement the Quality Evaluation framework
Build and maintain the Quality Evaluation Agent responsible for multi-layer quality assessment across content segments, files, and cross-file outputs. Implement the Quality × Confidence scoring model and ensure iterative improvement loops operate within defined token and compute budgets. Integrate adversarial “Red Team” validation layers to test output robustness and guard against failure cases.

5. Ensure reliability, observability, and performance of the agent system
Monitor and optimize the overall agent pipeline for reliability, performance, and transparency. Implement structured logging, metrics, and tracing to enable clear debugging, system health monitoring, and continuous improvement of agent behaviors.

Ideal Experience:

• Python - 3.12+, Async/Await, FastAPI.

• LangGraph / LangChain - Practical experience building multi-step agent workflows, Not just tutorials.

• Production service patterns - Docker, health checks, structured logging, graceful shutdown, circuit breakers.

• Testing agentic systems - deterministic testing strategies for non-deterministic outputs, mock LLM responses, integration test patterns.

• Message queues - RabbitMQ consumer/producer patterns for async agent coordination.

Nice to have:

• AWS Bedrock - Familiarity with translation/NLP domain, experience with multi-model routing.