GitHub Copilot
AI coding assistant named explicitly in the brief for accelerating software delivery and developer productivity.
Use AI coding tools inside normal engineering controls: scope, tests, review and release.
Claude Code
Agentic coding environment for repo-aware analysis, implementation, testing, refactoring and engineering automation.
Use AI coding tools inside normal engineering controls: scope, tests, review and release.
LLMs
Large Language Models used as the reasoning and generation layer inside production AI applications.
Treat prompts and model choice as testable production inputs, not one-off chat experiments.
Generative AI
Applications that generate, transform, reason over or act on content using foundation models.
Treat prompts and model choice as testable production inputs, not one-off chat experiments.
OpenAI GPT
OpenAI foundation models used through APIs for reasoning, generation and application workflows.
Treat prompts and model choice as testable production inputs, not one-off chat experiments.
Azure OpenAI
Enterprise Azure-hosted access to OpenAI models with Microsoft cloud governance and integration.
Treat prompts and model choice as testable production inputs, not one-off chat experiments.
Anthropic Claude
Anthropic model family used for reasoning, coding, agents and enterprise AI applications.
Treat prompts and model choice as testable production inputs, not one-off chat experiments.
Claude APIs
Programmatic integration of Claude models into services, applications and agent workflows.
Treat prompts and model choice as testable production inputs, not one-off chat experiments.
Prompt Engineering
Designing instructions, context and output constraints so models behave reliably for a defined task.
Treat prompts and model choice as testable production inputs, not one-off chat experiments.
Prompt Optimisation
Iteratively improving prompts against measurable quality, cost, latency and failure criteria.
Treat prompts and model choice as testable production inputs, not one-off chat experiments.
AI Agents
LLM-powered components that reason, select tools and execute bounded multi-step tasks.
Agents should plan, use tools and automate work while remaining observable, testable and controlled.
Agentic AI
Systems where models can plan and take controlled actions rather than only return text.
Agents should plan, use tools and automate work while remaining observable, testable and controlled.
Copilots
AI assistants embedded into human workflows to accelerate decisions, coding or operational tasks.
Agents should plan, use tools and automate work while remaining observable, testable and controlled.
Workflow Automation
Connecting AI reasoning to deterministic business and engineering actions through tools and APIs.
Agents should plan, use tools and automate work while remaining observable, testable and controlled.
LangChain
Framework for composing LLM calls, retrieval, tools and application workflows.
Agents should plan, use tools and automate work while remaining observable, testable and controlled.
LangGraph
Graph-based orchestration framework for stateful, multi-step and agentic LLM workflows.
Agents should plan, use tools and automate work while remaining observable, testable and controlled.
CrewAI
Multi-agent framework for assigning roles and coordinating collaborative AI tasks.
Agents should plan, use tools and automate work while remaining observable, testable and controlled.
AutoGen
Microsoft-originated framework for conversational and multi-agent orchestration patterns.
Agents should plan, use tools and automate work while remaining observable, testable and controlled.
Semantic Kernel
Microsoft SDK for integrating models, plugins, memory and orchestration into applications.
Agents should plan, use tools and automate work while remaining observable, testable and controlled.
Code Generation
Using AI to produce or modify production code within normal engineering quality controls.
Agents should plan, use tools and automate work while remaining observable, testable and controlled.
AI-assisted SDLC
Applying AI across analysis, coding, tests, review, documentation, release and maintenance.
Agents should plan, use tools and automate work while remaining observable, testable and controlled.
Developer Productivity
Reducing engineering cycle time while protecting quality, maintainability and operational safety.
Agents should plan, use tools and automate work while remaining observable, testable and controlled.
Python
Primary language for AI services, orchestration, data processing and backend application development.
The AI layer still depends on strong APIs, languages, architecture and scalable service boundaries.
FastAPI
Python web framework commonly used for typed, high-performance AI and microservice APIs.
The AI layer still depends on strong APIs, languages, architecture and scalable service boundaries.
JavaScript
Core web and application language named as an acceptable software-engineering skill in the brief.
The AI layer still depends on strong APIs, languages, architecture and scalable service boundaries.
TypeScript
Typed JavaScript used for maintainable frontend, Node.js and full-stack production systems.
The AI layer still depends on strong APIs, languages, architecture and scalable service boundaries.
C#
Enterprise application language explicitly accepted by the role alongside Python, JavaScript/TypeScript and Java.
The AI layer still depends on strong APIs, languages, architecture and scalable service boundaries.
Java
Enterprise backend language explicitly accepted by the role for production application engineering.
The AI layer still depends on strong APIs, languages, architecture and scalable service boundaries.
REST APIs
HTTP interfaces used to expose AI capabilities and connect models with enterprise systems.
The AI layer still depends on strong APIs, languages, architecture and scalable service boundaries.
Microservices
Independently deployable service boundaries used to integrate AI safely into larger platforms.
The AI layer still depends on strong APIs, languages, architecture and scalable service boundaries.
Scalable Architecture
Designing applications so model, data, API and infrastructure layers can scale independently.
The AI layer still depends on strong APIs, languages, architecture and scalable service boundaries.
Cloud-native
Building services around managed cloud infrastructure, automation, observability and elastic deployment.
The AI layer still depends on strong APIs, languages, architecture and scalable service boundaries.
Enterprise Integration
Connecting AI to real identity, APIs, data, workflows and systems of record rather than isolated demos.
The AI layer still depends on strong APIs, languages, architecture and scalable service boundaries.
RAG
Retrieval-Augmented Generation: grounding model responses in retrieved enterprise knowledge.
Retrieval quality, permissions and provenance determine whether enterprise AI can be trusted.
Embeddings
Vector representations of text or data used for semantic retrieval and similarity search.
Retrieval quality, permissions and provenance determine whether enterprise AI can be trusted.
Semantic Search
Retrieving information by meaning rather than only exact keyword matching.
Retrieval quality, permissions and provenance determine whether enterprise AI can be trusted.
Vector Database
Storage and search layer optimised for embeddings and similarity-based retrieval.
Retrieval quality, permissions and provenance determine whether enterprise AI can be trusted.
Pinecone
Managed vector database named in the job description for RAG and semantic retrieval.
Retrieval quality, permissions and provenance determine whether enterprise AI can be trusted.
ChromaDB
Open-source embedding database named in the brief for vector search and RAG workflows.
Retrieval quality, permissions and provenance determine whether enterprise AI can be trusted.
Weaviate
Vector database and semantic-search platform explicitly listed in the role requirements.
Retrieval quality, permissions and provenance determine whether enterprise AI can be trusted.
Azure AI Search
Azure search service supporting enterprise retrieval, hybrid search and vector search.
Retrieval quality, permissions and provenance determine whether enterprise AI can be trusted.
Knowledge Repositories
Authoritative document, database and content sources used to ground AI applications.
Retrieval quality, permissions and provenance determine whether enterprise AI can be trusted.
Model Evaluation
Testing model outputs against task-specific quality, correctness and reliability criteria.
Quality and safety must be measured through evals, traces, monitoring, policy and human oversight.
LLM Evals
Repeatable automated or human evaluation sets used to detect regressions in AI behaviour.
Quality and safety must be measured through evals, traces, monitoring, policy and human oversight.
Observability
Tracing model calls, retrieval, tools, errors, latency and cost across production AI workflows.
Quality and safety must be measured through evals, traces, monitoring, policy and human oversight.
Model Monitoring
Watching production AI quality, drift, failures, latency, usage and operational health over time.
Quality and safety must be measured through evals, traces, monitoring, policy and human oversight.
AI Governance
Policies and controls defining acceptable AI use, accountability, auditability and change management.
Quality and safety must be measured through evals, traces, monitoring, policy and human oversight.
Responsible AI
Designing AI systems around safety, transparency, privacy, fairness and accountable human oversight.
Quality and safety must be measured through evals, traces, monitoring, policy and human oversight.
Security
Protecting identities, secrets, data, tools and model boundaries with explicit access controls.
Quality and safety must be measured through evals, traces, monitoring, policy and human oversight.
Human-in-the-loop
Requiring human review or approval for higher-risk AI decisions and actions.
Quality and safety must be measured through evals, traces, monitoring, policy and human oversight.
AWS
Cloud platform listed as desirable for deploying and operating production AI applications.
Enterprise AI has to survive identity, cloud, CI/CD, containers, stakeholders and real operational constraints.
Azure
Microsoft cloud platform relevant to Azure OpenAI, enterprise identity, DevOps and AI services.
Enterprise AI has to survive identity, cloud, CI/CD, containers, stakeholders and real operational constraints.
GCP
Google Cloud Platform, included among the desirable cloud-platform skills in the brief.
Enterprise AI has to survive identity, cloud, CI/CD, containers, stakeholders and real operational constraints.
GitHub Actions
CI/CD automation for testing, validating and releasing software directly from GitHub workflows.
Enterprise AI has to survive identity, cloud, CI/CD, containers, stakeholders and real operational constraints.
Azure DevOps
Microsoft platform for repos, pipelines, work tracking and enterprise software delivery.
Enterprise AI has to survive identity, cloud, CI/CD, containers, stakeholders and real operational constraints.
Jenkins
CI/CD automation server listed as a desirable engineering-delivery technology.
Enterprise AI has to survive identity, cloud, CI/CD, containers, stakeholders and real operational constraints.
Docker
Container technology for packaging AI services and their dependencies consistently across environments.
Enterprise AI has to survive identity, cloud, CI/CD, containers, stakeholders and real operational constraints.
Kubernetes
Container orchestration platform for scheduling, scaling and operating distributed services.
Enterprise AI has to survive identity, cloud, CI/CD, containers, stakeholders and real operational constraints.
Terraform
Infrastructure-as-code tooling for reproducible cloud environments and platform configuration.
Enterprise AI has to survive identity, cloud, CI/CD, containers, stakeholders and real operational constraints.
CI/CD
Automated build, test, security and deployment controls that move AI code safely into production.
Enterprise AI has to survive identity, cloud, CI/CD, containers, stakeholders and real operational constraints.
APIs
The service contracts through which AI capabilities, enterprise data and tools are integrated.
Enterprise AI has to survive identity, cloud, CI/CD, containers, stakeholders and real operational constraints.
Mentoring
Helping developers adopt AI-assisted engineering patterns rather than creating dependence on one specialist.
Enterprise AI has to survive identity, cloud, CI/CD, containers, stakeholders and real operational constraints.
Stakeholder Use Cases
Turning business needs into prioritised AI opportunities with measurable outcomes and delivery constraints.
Enterprise AI has to survive identity, cloud, CI/CD, containers, stakeholders and real operational constraints.
Enterprise AI
Production AI built within real organisational constraints: data, identity, governance, reliability and operations.
Enterprise AI has to survive identity, cloud, CI/CD, containers, stakeholders and real operational constraints.
MCP
Model Context Protocol: a standard way to expose typed tools and context to coding agents and AI applications.
Reusable tools, MCP, state and delegation turn an LLM into a controlled engineering agent.
Tool Calling
Allowing a model or agent to select and invoke approved external functions, APIs or services.
Reusable tools, MCP, state and delegation turn an LLM into a controlled engineering agent.
Function Calling
Structured model output that maps natural-language intent into deterministic function arguments.
Reusable tools, MCP, state and delegation turn an LLM into a controlled engineering agent.
Multi-agent Systems
Architectures where specialised agents coordinate through explicit roles, hand-offs and shared state.
Reusable tools, MCP, state and delegation turn an LLM into a controlled engineering agent.
Router / Planner / Executor
Common agent pattern that separates task classification, decomposition and action execution.
Reusable tools, MCP, state and delegation turn an LLM into a controlled engineering agent.
Memory + State
Durable conversation or workflow state that lets agents continue tasks safely across multiple steps.
Reusable tools, MCP, state and delegation turn an LLM into a controlled engineering agent.
Subagents
Specialised child agents delegated bounded parts of a larger engineering or reasoning task.
Reusable tools, MCP, state and delegation turn an LLM into a controlled engineering agent.
Skills
Reusable packaged instructions, tools and workflows that make AI-assisted engineering repeatable.
Reusable tools, MCP, state and delegation turn an LLM into a controlled engineering agent.
Hooks
Lifecycle automations that run before or after agent actions, tests, commits or deployment events.
Reusable tools, MCP, state and delegation turn an LLM into a controlled engineering agent.
Guardrails
Deterministic checks that constrain model inputs, outputs or actions according to policy.
Production AI needs deterministic controls around probabilistic behaviour: security, traces, retries and cost.
Prompt Injection
Adversarial instructions hidden in user or retrieved content that try to override trusted system intent.
Production AI needs deterministic controls around probabilistic behaviour: security, traces, retries and cost.
PII Redaction
Detecting and removing personally identifiable information before inappropriate exposure or model use.
Production AI needs deterministic controls around probabilistic behaviour: security, traces, retries and cost.
Secrets Management
Keeping API keys, credentials and tokens outside prompts and source code with controlled access.
Production AI needs deterministic controls around probabilistic behaviour: security, traces, retries and cost.
Least Privilege
Giving users and agents only the minimum permissions required for the current task.
Production AI needs deterministic controls around probabilistic behaviour: security, traces, retries and cost.
Audit Trail
A durable record of model, retrieval, tool, approval and deployment decisions for accountability.
Production AI needs deterministic controls around probabilistic behaviour: security, traces, retries and cost.
Groundedness
The degree to which an AI answer is supported by authorised source evidence rather than unsupported generation.
Production AI needs deterministic controls around probabilistic behaviour: security, traces, retries and cost.
Hallucination
Unsupported or fabricated model output that must be reduced through retrieval, constraints and evaluation.
Production AI needs deterministic controls around probabilistic behaviour: security, traces, retries and cost.
Golden Dataset
A curated set of representative inputs and expected outcomes used to evaluate AI quality consistently.
Production AI needs deterministic controls around probabilistic behaviour: security, traces, retries and cost.
Regression Testing
Checking that prompt, model, retrieval or code changes do not break previously working behaviour.
Production AI needs deterministic controls around probabilistic behaviour: security, traces, retries and cost.
Tracing
Following a request across model calls, retrieval, tools and services to diagnose behaviour and failures.
Production AI needs deterministic controls around probabilistic behaviour: security, traces, retries and cost.
Latency
End-to-end response time, a key production constraint when choosing models, retrieval and tool paths.
Production AI needs deterministic controls around probabilistic behaviour: security, traces, retries and cost.
Token Cost
The model-usage cost created by input and output tokens, context size and repeated calls.
Production AI needs deterministic controls around probabilistic behaviour: security, traces, retries and cost.
Model Routing
Selecting a model dynamically based on capability, cost, latency, policy or availability.
Production AI needs deterministic controls around probabilistic behaviour: security, traces, retries and cost.
Provider Fallbacks
Switching to an alternate model or provider when the preferred path is unavailable or unsuitable.
Production AI needs deterministic controls around probabilistic behaviour: security, traces, retries and cost.
Retries
Controlled re-execution of transiently failed calls with limits, backoff and idempotency.
Production AI needs deterministic controls around probabilistic behaviour: security, traces, retries and cost.
Caching
Reusing safe prior results or computed context to reduce latency, cost and repeated work.
Production AI needs deterministic controls around probabilistic behaviour: security, traces, retries and cost.
Idempotency
Designing repeated tool or API calls so retries do not create duplicate side effects.
Production AI needs deterministic controls around probabilistic behaviour: security, traces, retries and cost.
API Contracts
Explicit request, response, error and version schemas that separate AI reasoning from deterministic services.
Production AI needs deterministic controls around probabilistic behaviour: security, traces, retries and cost.
Schema Validation
Checking structured inputs and outputs against types or schemas before downstream use.
Production AI needs deterministic controls around probabilistic behaviour: security, traces, retries and cost.
OpenTelemetry
Open standard for collecting traces, metrics and logs across distributed production services.
Production AI needs deterministic controls around probabilistic behaviour: security, traces, retries and cost.
Git
Version-control foundation for safe branching, review, traceability and rollback of software changes.
AI-generated work should improve—not bypass—the review, testing, architecture and leadership disciplines of software engineering.
Pull Requests
Review boundary for inspecting AI- or human-generated code before it enters protected branches.
AI-generated work should improve—not bypass—the review, testing, architecture and leadership disciplines of software engineering.
Unit Testing
Fast tests for isolated functions and components used to catch deterministic software regressions.
AI-generated work should improve—not bypass—the review, testing, architecture and leadership disciplines of software engineering.
Integration Testing
Tests that verify models, APIs, databases and services work together across real boundaries.
AI-generated work should improve—not bypass—the review, testing, architecture and leadership disciplines of software engineering.
Architecture Decision Records
Short versioned records explaining important architecture choices, trade-offs and consequences.
AI-generated work should improve—not bypass—the review, testing, architecture and leadership disciplines of software engineering.
Technical Leadership
Setting engineering direction while remaining close enough to implementation to unblock teams and raise standards.
AI-generated work should improve—not bypass—the review, testing, architecture and leadership disciplines of software engineering.