Meet Larkfield
Larkfield is the online shop that every course uses as its example. It is made up. On this page, you meet its people, follow its AI work, and see where the data of each course comes from.
Who Larkfield is
A home and garden shop
Larkfield is a Canadian online shop for home and garden products. It also sells its own brand, Larkfield Basics. Marketplace sellers on its website add other categories, such as clothing.
Canada and the EU
Its head office is in Montréal. It sells to customers across Canada and in the European Union. Prices are in Canadian dollars, and customers in the EU pay in euros.
Two languages
Customers write to Larkfield in English and in French. The help desk has bilingual agents. Larkfield built its help-desk software itself.
Online, then shops too
Larkfield sold only online until 2026. That year, it buys a chain of garden centres and a hardware chain. It also has an outlet shop for second-hand items.
A second shop, the bookshop Bramble Books, also uses Larkfield’s help desk.
The people
You meet the same people in many courses. Each person has one role in every course. People in the case studies work at other organisations, even when they have the same first name.
Larkfield’s AI work, step by step
The steps follow the learning paths. Each course tells its part of the story in its own time, so this page gives the order of events, not dates.
Find where AI can help
The help-desk team lists the problems that AI can help with, such as policy questions and sorting tickets. They learn to read the numbers that come with AI.
Learn from the help-desk data
Mei prepares the ticket data. Priya builds a model that finds the tickets that need a person, and the team puts it into service. A test with deep learning shows that a simpler model is good enough for now.
Use AI across the shop
Other teams use machine learning too: to find unusual payments, forecast demand, check returned items in photos, route messages, recommend and search products, match customer records after Larkfield buys its shops, test changes with experiments and understand phone calls. Each of these courses stands alone.
- Forecasting and Time Series: Predict What Comes Next
- Anomaly Detection: Find What Is Unusual
- Vision AI: See, Count and Inspect
- Text Classification and Routing: Send Each Message to the Right Place
- Recommendation and Personalisation: Show Each Person What Matters
- Search and Ranking: Find the Best Match
- Entity Resolution and Record Matching: Is This the Same Thing?
- Experiments and Causal Impact: Did Our Change Work?
- Speech AI: Transcribe, Understand and Speak
Draft replies with a language model
The team chooses a language model. Amira builds an assistant that drafts replies, and an agent checks and sends each one. The assistant then gets Larkfield’s policy documents, and a workflow that can act, with approvals. Priya measures each change.
Build and protect the platform
The help-desk software gets a real database, a plan for busy days such as Black Friday, a new screen for agents and sign-in. Bramble Books starts to use the help desk. The team checks the assistant for security risks and connects it to tools.
Run AI in production
The team tests a fine-tuned model and reads receipts and other documents with AI. It runs releases with checks, plans costs, watches the live workflow and controls all of its AI with a gateway and a registry. Leaders decide go or no go, and a governance board sets the rules.
- MLOps and LLMOps: Operating AI in Production
- Fine-Tuning and Adapting AI Models
- Document and Multimodal AI
- AI System Design and Cost Engineering
- Governing AI Agents: Gateways, Registries and Controls
- Observability for AI Agents: Traces, Costs and Audit Trails
- Leading AI Products and Teams
- AI Governance: Policies, Risk and Accountability
- Portfolio Capstone: Build, Evaluate and Ship an AI Product
The data
Each course has its own slice of Larkfield’s data. A course uses only the data that its lessons need, for its own period. So the numbers differ between courses: one course has 1,417 tickets, and another has 300,000. Each number is correct for its course.
Synthetic data is made up for the course. Adapted data is real public data that we changed to fit Larkfield. All the customers in the data are made up.
| Course | Data | Source | Period | Main sizes | Licence |
|---|---|---|---|---|---|
| AI Career Foundations: Roles, Skills and RoadmapsData lesson | Small examples inside the lessons, such as a ten-row ticket table | Synthetic | Not given | 10 tickets in the largest table | No data files |
| Essential Mathematics and Statistics for AIData lesson | Numbers inside the lessons: test tickets and delivery orders | Synthetic | Not given | Test sets of 300 and 1,000 tickets | No data files |
| SQL and Data Preparation for AIData lesson | The help-desk export | Synthetic | January to June 2026 | 240 customers, 1,417 tickets, 1,874 outcomes | |
| Machine Learning: From Problem to Reliable ModelData lesson | The full ticket export, with the tickets that went to a person | Synthetic | July 2024 to June 2026 | About 1,000 customers; 16,939 training, 1,179 validation and 1,002 test tickets | Not stated |
| Deep Learning and Transformers ExplainedData lesson | Ticket texts with their team, written by a language model from a list of situations | Synthetic | Not given | 5,896 tickets (4,696 / 600 / 600) | |
| Forecasting and Time Series: Predict What Comes NextData lesson | Daily demand, orders, tickets, staff rosters, weather and promotions | Synthetic | 1 April 2024 to 29 March 2026 | 200 products, 403,758 demand rows | |
| Anomaly Detection: Find What Is UnusualData lesson | Card payments, refunds, orders, tickets and ticket-API metrics, with planted anomalies | Synthetic | Payments: June to November 2025. Orders: April 2024 to March 2026 | 343,074 payments, 24,038 customers | |
| Vision AI: See, Count and InspectData lesson | Catalogue photos, photos of returned items, warehouse scenes and customer photos | Synthetic and real public data | Until April 2026 | 733 catalogue photos, 735 return photos; 2,113 files | |
| LLMs: How They Work and How to Choose OneData lesson | A small test set of tickets, in English and French, and the support policy | Synthetic | Not given | 24 tickets; 192 scored replies | Not stated |
| Building Reliable Applications with LLM APIsData lesson | Tickets and an orders database, with recorded model answers | Synthetic | Not given | 69 tickets; 276 recorded answers | |
| Deploying AI Models for Real UsersData lesson | The tickets of the machine learning course, plus one new month | Synthetic | July 2024 to July 2026 | 18,118 training tickets; 1,400 new tickets in July | Not stated |
| Databases and Storage for AI ApplicationsData lesson | Help-desk data in a small size and a large size | Synthetic | Small: 25 September to 8 October 2026. Large: one year to 8 October 2026 | Small: 40 customers, 200 tickets. Large: 20,000 customers, 300,000 tickets | |
| RAG: Building AI That Uses Your DataData lesson | The policy collection and Grace’s test questions | Synthetic | Not given | 36 documents, 185 passages, 67 questions | |
| Evaluating and Testing AI SystemsData lesson | An evaluation set, and one simulated month of tickets | Synthetic | Not given | 271 cases (102 / 98 / 71); 1,270 simulated tickets | |
| Testing AI Applications: Manual and AI-Assisted QAData lesson | A practice help-desk app with test data, and recorded model replies | Synthetic | Not given | 4 customers, 8 orders, 8 reviews; 300 recorded replies | |
| AI Agents and Workflow OrchestrationData lesson | Workflow tasks, a practice database and policy passages | Synthetic | Not given | 70 tasks, 24 policy passages | |
| AI-Assisted Software DevelopmentData lesson | A ticket file and a web-form export in a small code project | Synthetic | Not given | 15 tickets | Not stated |
| Frontend Architecture and Performance for AI AppsData lesson | A mock ticket API, with replayed AI drafts | Synthetic | Not given | 10,000 tickets; 69 recorded drafts | |
| Authentication, Permissions and Multi-Tenant ApplicationsData lesson | The help-desk data of the databases course, plus a second shop, Bramble Books | Synthetic | 25 September to 8 October 2026 | Larkfield: 40 customers, 200 tickets. Bramble Books: 12 customers, 40 tickets | |
| AI Security, Privacy and Responsible DesignData lesson | A practice support assistant: database, documents and web pages, with planned attacks | Synthetic | Not given | 26 normal tasks, 42 attacks, 31 passages | |
| MCP: Connect AI Applications to Tools and DataData lesson | Tickets and policy documents of two shops on one help desk | Synthetic | Not given | 21 tickets, 14 documents | |
| Scaling APIs and AI WorkloadsData lesson | The large help-desk data of the databases course, with recorded provider timings | Synthetic | 9 October 2025 to 8 October 2026 | 300,000 tickets, 20,000 customers | |
| MLOps and LLMOps: Operating AI in ProductionData lesson | The evaluation course’s tickets, recorded model calls and saved service data | Synthetic | Not given | 271 tickets; 1,358 recorded calls; a practice week of 5,754 tickets | |
| Fine-Tuning and Adapting AI ModelsData lesson | The deep learning course’s training tickets, with new labels | Synthetic | Not given | 4,696 tickets; 900 in the training sample | |
| Document and Multimodal AIData lesson | A document set: till receipts, policies, a price list and problem files | Synthetic | Receipts from June to September 2026 | 65 documents, 53 of them receipts | |
| AI System Design and Cost EngineeringData lesson | The RAG course’s documents and questions, a demand file and measured costs | Synthetic | Not given | 36 documents, 67 questions, 659 recorded calls | |
| Governing AI Agents: Gateways, Registries and ControlsData lesson | A provider simulator: recorded answers from earlier courses and a simulated bill | Synthetic | One month: September 2026 | 5,253 recorded answers; 40,754 model calls | |
| Observability for AI Agents: Traces, Costs and Audit TrailsData lesson | One saved week of the support workflow: simulated traffic and replayed decisions | Synthetic | 12 to 18 October 2026 | 316 runs, 1,331 audit records | |
| Leading AI Products and TeamsData lesson | An evidence pack, with numbers from the evaluation, operations and document courses | Synthetic | Not given | 271 tickets; 199 judged drafts | |
| AI Governance: Policies, Risk and AccountabilityData lesson | The AI portfolio, the policy and the governance board’s files | Synthetic | Not given | 8 AI systems | |
| Portfolio Capstone: Build, Evaluate and Ship an AI ProductData lesson | The RAG course’s documents, plus Bramble Books’ documents and a question bank | Synthetic | Not given | 36 + 15 documents; 103 questions | |
| Text Classification and Routing: Send Each Message to the Right PlaceData lesson | The support inbox, in English and French | Synthetic | September 2025 to January 2026 | 10,000 messages (7,071 English, 2,929 French) | |
| Recommendation and Personalisation: Show Each Person What MattersData lesson | The shop log: views, basket adds and orders | Synthetic | 1 February 2025 to 5 July 2026 | 2,994 products, 40,000 customers, 805,316 events | |
| Search and Ranking: Find the Best MatchData lesson | The catalogue, test queries and judgments from a real public dataset (WANDS); prices, stock and the query log are made up | Adapted from real public data | Query log: 1 September to 26 October 2025 | 40,066 products, 624 queries, 20,000 searches | |
| Entity Resolution and Record Matching: Is This the Same Thing?Data lesson | Customer lists of the shop, the loyalty programme and the garden centres, and two supplier feeds | Synthetic | Not given | 60,108 records of 37,500 people; supplier feeds of 4,460 and 3,501 products | |
| Experiments and Causal Impact: Did Our Change Work?Data lesson | A simulator of the shop, its customers and their orders | Synthetic | June 2024 to November 2025 | 40,000 to 93,463 customers in each scenario | |
| Speech AI: Transcribe, Understand and SpeakData lesson | Phone calls to the help desk, made with open text-to-speech voices | Synthetic | Not given | 340 calls (260 for learners: 175 English, 85 French) |
These courses use a simple help desk and do not name Larkfield: Python for Practical AI Engineering; Git and Team Development Essentials; Build and Deploy Your First API; Docker and Cloud Essentials for AI Developers; CI/CD, Testing and Safe Releases.
Is Larkfield real?
No. Larkfield, its people and its customers are made up. A real company or person with the same name is a coincidence.
Why one shared company helps you
- You learn the business once. Each new course starts with a problem that you already know.
- You see how the parts connect: the model of one course becomes the data, the cost or the risk of a later course.
- The decisions feel real: costs, customers, rules, and people who do not agree.
- No real customer’s data is used, so we can show you all of it.