Become a Forward Deployed
AI Engineer.

• Interview stakeholders; map the workflow, decisions, exceptions and human handoffs.
• Compare business value, delivery effort and risk; decide where AI or simpler automation fits.
• Agree the users, scope, data access and responsibilities with the client.
• Define acceptance criteria, baseline metrics and a testable delivery plan for your capstone.

Your deliverableClient brief + process map + acceptance criteria

• Map documents, databases and APIs; verify ownership, access rights and sensitive fields.
• Check missing values, duplicates, bias and freshness; document data limitations.
• Build a repeatable ingestion pipeline with source references and update handling.
• Create a held-out evaluation set covering normal requests, edge cases and unsafe requests.

Your deliverableRepeatable data pipeline + access map + evaluation dataset

• Map the user journey and the decisions the agent may make or must escalate.
• Prototype the key screens, tool actions and approval steps with AI-assisted tools.
• Make sources, uncertainty, errors and human takeover clear in the interface.
• Test realistic tasks with users; turn feedback into a prioritised implementation backlog.

Your deliverableTestable prototype + human handoff flow + user feedback

• Compare model capabilities, context limits, latency and cost against your client brief.
• Build with RAG and tool calls; assess when multiple agents add value.
• Define orchestration, stop conditions and fallback behaviour for multi-step tasks.
• Run your evaluation set; compare answer quality, tool-use success, speed and cost per task.

Your deliverableWorking agent + evaluation results + architecture decision record

• Structure reusable agent, tool and API components; document contracts and dependencies.
• Connect business systems with authentication, validation, timeouts and safe retries.
• Use AI coding assistants with Git, code review and automated integration tests.
• Package the application with environment configuration, CI/CD and a tested rollback path.

Your deliverableIntegrated agent + reusable components + tests + deployment pipeline

• Test prompt injection, data leakage and unsafe tool actions; enforce least privilege.
• Trace agent behaviour; measure quality, failures, latency and cost.
• Load-test the workflow; compare caching, queues and model routing against a baseline.
• Set token budgets and operating limits; test alerts, fallbacks and recovery procedures.

Your deliverableSecurity test results + load report + cost and reliability dashboard

• Define autonomy limits, human checkpoints and escalation rules.
• Assign owners; document risks, controls and change reviews with ISO/IEC 42001 as a reference.
• Use evaluation and security evidence to make a go/no-go recommendation for rollout.
• Prepare a reusable runbook; rehearse client handover, incident response and user onboarding.

Your deliverableGovernance dossier + release checklist + client runbook