AI Native Builder

Forward Deployed Engineer

Location:

Bristol, UK | Type:
Contract

We are hiring Forward Deployed Engineers who treat AI as the substrate of how software gets built, not a tool to be cautious of, but the medium they work in.

• Sit with a business or clinical leader and reframe an idea or problem into something concrete and buildable.
• Know what to build by the end of the conversation; have a working prototype to react to by the end of the week.
• Partner closely with product, design, and client stakeholders to translate ambiguous ideas into software that ships.
• Demo live without a slide deck. Reframe problems out loud. Don't get stuck waiting for someone else to make the decision.
• Lead POCs, innovation sprints, and internal research experiments to validate emerging AI techniques.

• When the brief is clear, head down and produce.
• Build modular backends in Python or Type Script aligned with clean architecture, OOP, SOLID, and domain‑driven design.
• Create fullstack applications, APIs, agents, workflows, and similar systems using frameworks such as Next.js, React, FastAPI, Fastify, FastMCP, and Hono.
• Architect and ship production‑grade agentic applications using Lang Graph, Auto Gen, Claude Agent SDK, OpenAI Assistants, or your own orchestration layer.
• Integrate frontier and self‑hosted LLMs (Claude, GPT, Gemini, open‑weight models) with tools, data, and external systems through MCP and custom connectors.
• Apply RAG techniques where they actually help: vector databases (Pinecone, Chroma, Weaviate, pgvector), hybrid retrieval with Elastic Search or Solr, and BM25 + similarity search.
• Work across relational, document, key‑value, and graph stores as the problem demands; use event‑driven patterns where they fit, not by default.
• Design prompt and context engineering frameworks that optimize accuracy, repeatability, cost, and latency.
• Use AI‑assisted development tools (Claude Code, Git Hub Copilot, Cursor, Codex) through structured workflows, native instructions, templates, and sub‑agents with discipline and review.
• Fine‑tune or adapt models where the problem genuinely calls for it.

• Spin up the infra, write the evals, wire up the MCP servers, deploy the agents, and harden the bits that survive contact with real users.
• Deploy on AWS, Azure, Cloudflare, or Vercel using containerization (Docker, Kubernetes) or serverless chosen for fit, not preference.
• Treat evals as a first‑class discipline: hands‑on harnesses, not theoretical frameworks. Build with a clear‑eyed view of where current AI tooling helps and where it falls short.
• Apply engineering practices that hold up in production: TDD, secrets management and rotation, SAST/DAST, structured logging, metrics, tracing, and automated CI/CD (Git Hub Actions, Jenkins).
• Own what you build end‑to‑end, including the infrastructure and operations that keep it running.
• Mentor others on system design, agentic patterns, and AI engineering best practices.

• 3+ years relevant experience building production applications using AI / agentic development approaches-fullstack applications, agents, workflows, MCPs, and more.
• Hands‑on experience with agents, not just prompted models. You have wired tools to a model and let it run multi‑step using Lang Graph, Auto Gen, Claude Agent SDK, OpenAI Assistants, or your own orchestration.
• Active, structured use of AI‑assisted development tools (Claude Code, Cursor, Git Hub Copilot) with demonstrable workflows, sub‑agents, skills, and innovative approaches.
• Strong Python or Type Script, with OOP, SOLID, 12‑factor application development, and microservice architecture. You've built Next.js applications, FastAPI services, and similar.
• End‑to‑end implementation experience with vector databases, retrieval pipelines, and eval harnesses.
• Cloud‑native deployment experience across at least one of AWS, Azure, Cloudflare, or Vercel‑with Docker, Kubernetes, and Git Hub Actions.
• A no‑compromise attitude on clean code, TDD, security, observability, scalability, performance, and cost.
• A deep working understanding of how LLMs behave‑and where they break‑and how to…