Module 18: Who uses this: sectors and careers · Lesson 18.11 · about 20 min
Local AI skills in job ads
The local AI skills employers ask for in data, software, IT and research roles, the words they use in job ads, and how to show them with projects from this course.
Module 18 · about 20 minutes · read first: any of Modules 9 to 15. You’ll be able to describe your local AI skills in the words employers use, with projects to prove them.
1. Why this matters
Many organisations want AI but cannot send data to outside services. They need people who can run models in-house: data analysts, developers, IT staff, researchers. The skills in this course match those job ads closely.
2. Words in job ads, and where you learned them
| Job ad phrase | What it means | Course |
|---|---|---|
| LLM, generative AI | Large language models | Modules 2 and 3 |
| On-premise, self-hosted, private AI | Running models on the organisation’s own hardware | Modules 5, 9, 15 |
| Ollama, llama.cpp, vLLM, LM Studio | Engines | Modules 9 and 15 |
| RAG, embeddings, vector database | Document search with AI | Module 12 |
| Agents, tool calling, function calling | Models that act with tools | Module 13 |
| Fine-tuning, LoRA | Teaching a model new habits | Module 14 |
| Quantisation, GGUF, inference optimisation | Making models fit and run fast | Module 8, 17.2, 17.3 |
| Prompt engineering | Writing instructions that work | Module 10 |
| Model evaluation | Testing models on real tasks | 12.7, 17.1 |
| AI governance, model risk | Rules, records, monitoring | 15.5, 17.7, 18.3 |
| Python, APIs, Docker, Linux | Tools around the models | Modules 11 and 15 |
3. Projects that prove it
Pick two or three, put the code on GitHub with a clear README (the front page that explains what a project does and how to run it), and write a short post about each:
- Private document search over a public document set (Mini projects in Module 12), with a test set and accuracy.
- SQL helper (mini project 8) against a sample database, with the questions it got right and wrong.
- Team server (mini project 7) documented as a setup guide with security choices.
- Cost study for a made-up company using the break-even calculator (Lesson 16.4) and a hybrid policy (Lesson 15.6).
- Capstone (Lesson 18.12).
4. Writing it on a CV
Lead with the result, then the tools: “Built a private document search over 2,000 policy PDFs with Ollama and a local vector index; 88% of test questions answered with correct citations.” Use only numbers you measured.
Start weak. "Used AI for SQL." True, but it says nothing a hiring manager can picture.
Add the action. "Built a local SQL assistant..." Now it is something you made.
Add the tools. "...with Gemma 4, Ollama and SQLite..." These are the words in the job ad.
Add a measured result. "...17 of 20 test questions correct on first try." Now it is evidence, and it starts a technical conversation in the interview.
Maths and principle
Principle: show, measure, explain.
Equation: a strong bullet = action + tool + measured result.
Analogy: a cooking audition: they taste the dish, not the recipe list.
Try it: turn “used AI for SQL” into a strong bullet. (For example: “Built a local SQL assistant with Gemma 4 and SQLite; 17 of 20 test questions correct on first try.”)
- action
- what you did, with a strong verb: built, cut, set up, tested
- tool
- the named tools, in the words job ads use (Ollama, RAG, Docker)
- measured result
- a number you measured yourself: accuracy, time saved, cost
- Physical meaning
- A hiring manager reads a CV in under a minute. A measured result is the part they remember and ask about, because it shows that the thing worked and that you know how to test it.
A percentage cut is often the clearest result. Divide by the old time, not the new one:
“A long list of models and tools shows I know local AI.” Names without results show only that you have read about them. One working project with a test set and an honest number beats ten logos. Employers ask what you built and how well it worked.
5. Worked scenario
Made-up candidate. A new graduate in data analytics adds two projects to GitHub: a document search over public government reports with a 50-question test set, and a cost study for a 20-person office. In interviews for data roles, both projects lead to long technical conversations, and one employer asks them to set up the same search for internal policies in the first month.
6. Practice
- Pick two projects from section 3 to build first.
- Write one CV bullet for a project you have done.
- Search a job board for “Ollama” or “RAG” in your city and note three skills that keep appearing.
Answers
- Your answer; document search is the most asked-for.
- Action + tool + measured result.
- Your answer; Python, RAG and Docker are common.
New words: on-premise, self-hosted, portfolio project, README.
- Job ads for in-house AI use words like on-premise, self-hosted, RAG, Ollama, vLLM, agents, LoRA and model evaluation.
- Each of those maps to a module of this course.
- Prove skills with two or three working projects on GitHub, each with a README, a test set and an honest result.
- A strong CV bullet = action + tool + measured result; use only numbers you measured.
Check yourself
Sources
- Job boards searched for terms such as "Ollama", "vLLM", "RAG", "LLM" and "on-premise AI" (Indeed, LinkedIn Jobs, Naukri, Job Bank), November 2026; counts change daily, so search them yourself
- Lessons in Modules 9 to 15 of this course
- World Economic Forum, The Future of Jobs Report 2025 (AI and big data among the fastest-growing skills), https://www.weforum.org/publications/the-future-of-jobs-report-2025/
- U.S. Bureau of Labor Statistics, Occupational Outlook Handbook: Software Developers (official page), https://www.bls.gov/ooh/computer-and-information-technology/software-developers.htm
Your ticks stay in this browser. Save them as a file on the course home.
Updated 2026-11-22. Prices and model names change fast: check the official source before you buy.