Build a system that resolves support tickets step by step, and stays safe while it does. You turn Larkfield's help desk work into a workflow that looks up evidence, proposes an action, waits for a person's approval before any change, and records what happened.
Read every lesson as a guest, or sign in to enroll and track your progress.
Enroll to save the lessons that you complete. It is free.
By the end of this course, you will be able to:
A language model can read a support ticket and suggest what to do. Letting it act is a different job. It must look up the right order, follow the policy, ask before it changes anything, and never refund the same charge twice when a service is slow to answer.
In this course, you build Larkfield’s support-resolution workflow step by step. You start with fixed steps that the code decides. Then you let the model choose steps only where that earns its place, and you measure whether it does. The task keeps an explicit state, every loop has a limit, a person approves every change, and every write can be repeated safely.
You also learn to check the result, not only the reply. The evaluation looks at what changed in the systems: the right action, nothing forbidden, and no duplicates.
You do not need an account or a paid key. The order, payment and shipping systems are a practice database on your computer, and every model decision has a recorded response from a real model. Diagrams and explorers carry the main ideas, so the text stays short.
You completed Building Reliable Applications with LLM APIs and Evaluating and Testing AI Systems, or you have the same skills. RAG is useful but not required. You do not need an API key or a paid account. Every model decision has a recorded response, every change goes to a practice database on your computer, and live calls with a free local model or your own key are optional.
Courses to take first: Building Reliable Applications with LLM APIs, Evaluating and Testing AI Systems.
People who have built an application on a model API and tested it, as in Building Reliable Applications with LLM APIs and Evaluating and Testing AI Systems, and now want a model to take several steps and use tools without losing control. Developers, data scientists and professionals moving into AI engineering.
You can study the ideas of this course without writing code. Read modules 1, 4 and 6 (about 7–9 hours), answer their knowledge checks, and get a free concepts certificate. You can come back for the practice later.
The paths for people who lead or use AI take this course this way. See the paths
Experienced AI practitioners designed this course. AI helped us draft the content; people chose the teaching approach and the examples, and reviewed every lesson. The outputs that you see are the actual results of running the code. How we make our courses
Choose how much the model decides. Run the same support task as fixed steps, as a router and as an agent, learn the common patterns, and measure when the simple design is the better one.
Give the task a state and the tools a contract. Keep goals, evidence, approvals and errors in typed fields, check every tool call and permission in code, and make the loop stop.
Plan with evidence and remember with care. Break a task into steps you can check, change the plan when the facts change, choose what the model sees, and decide what to keep about a returning customer.
Put a person before every change. Show the exact action for approval, save the task so that it can stop and resume, and make writes safe to repeat with operation IDs and reconciliation.
Split work only when it helps. Run independent lookups in parallel with clear merge rules, compare one agent with two workers on recorded runs, and see where tool integrations put the boundary.
Prove that it works before you trust it. Score tasks on outcome, side effects, cost and time, read traces to find failures, and test attacks and outages until the required checks pass.
Three end-to-end case studies on open data. Evaluate a customer-service agent on a public benchmark, test an agent against instructions hidden in tool results, and compare a workflow with an agent on a public service process. Complete at least one for the certificate; each one is a portfolio piece.
Complete at least one case study for the certificate. The others are optional. What each one covers
Use every module in one practice project. You collect evidence, and check it against the course outcomes.
Realistic projects from start to finish, on real data, with every output explained. Each one makes a portfolio piece. Complete at least one for the certificate; the others are optional.
Evaluate a customer-service agent on a public benchmark2 hr 30 min
Run an agent on public customer-service tasks with mock tools and a written policy, and score task success, policy violations and side effects over repeated runs.
Test an agent against instructions in tool results2 hr
Measure how often instructions hidden in tool results take over an agent, with and without defences, next to how often the agent still completes its task.
Workflow or agent? Automate an open-data process2 hr 30 min
Automate a public service process from open data twice, as a fixed workflow and as an agent, and compare errors, cost and time on the same tasks.
Loading the reviews…
Took this course? Sign in to review it. Learners can write a review after they complete 25% of the lessons.
