Loading...
Please sign in to view your progress
Gen AI interviews, from how LLMs work to RAG, evals and agents. Train with Coach Ada, then take a mock with Felix.
| # | Name | ||
|---|---|---|---|
1 | Start here. Twenty minutes with Felix: explain a concept, make a build call, say how it would fail. No hints on the call; your report scores five Gen AI signals and tells you which modules to do first.
| ||
2 | The Gen AI Interview, and How to Explain It Document | ||
3 | ![]() | Explain any Gen AI idea in three depths, then tell your own AI story with the check you ran.
Watch Ada work one example in her 3D lab, take a timed drill, and write your own rules.
| |
4 | A Gen AI experience screen with Felix: tell him about a real piece of work with AI, what went
wrong and how you checked, why an AI-first company, then make a call when time saved collides
with a quiet error. Written report after.
| ||
5 | How LLMs Work, in Plain Words Document | ||
6 | ![]() | Explain how an LLM works to a non-technical listener, then go deeper for an engineer. Watch Ada
move the next-word columns in her 3D lab, trace the risks of treating a model like a person, take
a timed drill and write your own rules.
| |
7 | A concept round with Felix: explain transformers and LLMs to a PM first, then answer the
engineer's follow-ups, handle a PM who wants the temperature at zero, and make a call when
your own advice backfires. Written report after.
| ||
8 | Training and Tuning: Prompt, Retrieve or Tune? Document | ||
9 | ![]() | Know how a model becomes an assistant, then pick prompt, retrieval or tuning and prove it. Watch Ada work one example in her 3D lab, choose your own case, take a timed drill and write your own rules.
| |
10 | A technique-choice round with Felix: prompt, retrieval or fine-tune for a made-up insurer, taken deep. Defend your diagnosis, handle a tuned model that leaks a customer's name, then make a call on your own plan. Written report after.
| ||
11 | Decoding and Context: The Dials and the Window Document | ||
12 | ![]() | Say what the window and the dials change, what they cost, and how you would test. Watch Ada work one example in her 3D lab, pick a technique for a large document, take a timed drill and write your own rules.
| |
13 | A round with Felix on the context window for a made-up bank's document tool, taken deep. Explain the effects, handle documents longer than any window, and call a trade-off between accuracy, cost and wait. Written report after.
| ||
14 | Prompt Engineering: Brief, Test, Change One Thing Document | ||
15 | ![]() | Fix a failing prompt one change at a time, then prove it with a test set. Watch Ada fix one in
her 3D lab, pick your own first block for a merchant bot, take a timed drill and write your own
rules.
| |
16 | A prompt engineering round with Felix. Explain your approach, then write a real prompt on the
notepad, hear what it produces, and fix it one change at a time. Defend how you would test it
and make a ship, hold or change call. Written report after.
| ||
17 | RAG: Look It Up, Then Answer Document | ||
18 | ![]() | Look the answer up first. Watch a question land among its neighbours in Ada's 3D lab, see where
retrieval goes wrong, pick how to cut a page into chunks, take a timed drill and write your own
rules.
| |
19 | A Gen AI design round with Felix: build semantic search over a product catalogue so a chat
assistant can use it. Choose what to index and what to filter, name how it fails, handle a
freshness curveball, and make a ship, hold or change call. Written report after.
| ||
20 | Hallucinations: Why They Happen and What You Do About Them Document | ||
21 | ![]() | Why a model sounds sure when it is wrong, and what you do about it. Watch Ada trace a made-up
case to its cause in her 3D lab, pick your own first cut on a confident-but-wrong model, take a
timed drill and write your own rules.
| |
22 | A Gen AI round with Felix on hallucinations in a deployed model. Define the failure, find the
cause, choose your defenses and their costs, adapt when a citation looks real but is not, and
call a trade-off between wrong answers and hand-offs. Written report after.
| ||
23 | Evaluating an LLM: How Do You Know It Works? Document | ||
24 | ![]() | Say how you'd know an LLM works: the test set, the grader and the bar to ship. Watch Ada work one
example in her 3D lab, pick how to measure an agent, take a timed drill and write your own rules.
| |
25 | An evaluation round with Felix: how would you know an answer engine is good? Define it, build the
test set, check your graders, then read his results sheet and decide whether to ship. Ends with a
trade-off between cost and a sensitive slice. Written report after.
| ||
26 | Securing an LLM System: How It Gets Attacked and What Holds Document | ||
27 | ![]() | Name how an AI system gets attacked and where the defense holds. Watch Ada hijack an inbox
assistant in her 3D lab, pick where to stop personal data leaving, take a timed drill and write
your own rules.
| |
28 | A design round with Felix: build a retrieval assistant over a customer's private documents, then
defend it. Who can see what, what hides inside a document, what a cache remembers. A revoked-access
curveball and a ship-or-hold call from a red-team result. Written report after.
| ||
29 | Responsible AI: Who Can It Hurt, and How Would You Know? Document | ||
30 | ![]() | Name who an AI system can hurt, and how you would know. Watch Ada work a résumé screener in her
3D lab, pick the layer that stops harmful output, take a timed drill and write your own rules.
| |
31 | A responsible-AI round with Felix: the ethical risks of an AI agent that acts in a hospital. Name who
is hurt, pick the risk to fix first, defend your safeguards, answer a "rubber stamp" curveball and
make a call when review costs wait time. Written report after.
| ||
32 | Agents and Tools: When the Model Acts Document | ||
33 | ![]() | Know when to use an agent, how it fails, and where the limits go. Watch Ada work a support agent in
her 3D lab, pick where an agent belongs in a workflow, take a timed drill and write your own rules.
| |
34 | An agent design round with Felix: build an AI agent that adapts to tasks it has never seen, keep it
under control, read a pilot results sheet and decide how much freedom it earns. Written report after.
| ||
35 | Shipping LLMs: Latency, Cost and the Rollout Document | ||
36 | ![]() | Find where an LLM's time and money go, then ship the fix in stages. Watch Ada work a chat product in
her 3D lab, pick when a batch should go, take a timed drill and write your own rules.
| |
37 | A serving round with Felix: design a ChatGPT-style assistant and defend the trade-offs between
speed, cost and quality. Size it, route it, cache it, roll it out, then handle a launch spike and a
router that saves money and loses quality. Written report after.
| ||
38 | AI Product Sense: Pick the Job, Design for the Failure Document | ||
39 | ![]() | Pick the job, test whether AI is the right tool, and design for the moment it is wrong. Watch Ada
build one in her 3D lab, decide what an assistant may do alone, take a timed drill and write your
own rules.
| |
40 | An AI product round with Felix: design a file upload feature for an AI chat app. Pick the person
and the job, say why AI and why now, design for the failure, adapt when users upload huge files,
and call a trade-off between uploads and cost. Written report after.
| ||
41 | Vibe Coding in an Interview: Build Small, Verify, Say What You Gave Up Document | ||
42 | ![]() | Build small with an AI tool, check what it made, and say what you gave up. Watch Ada review a diff
in her 3D lab, pick your own first move on a half-built codebase, take a timed drill and write your
own rules.
| |
43 | A live prototype round with Felix. Cut a thin slice, write a plan and a prompt, hear what the AI
tool produced line by line, and review the diff. Catch what it got wrong, handle a diff that
removes a test, and make a ship, tighten or rewrite call. Written report after.
| ||
44 | Company Rooms: What Each Loop Weighs Document | ||
45 | A practice room in the style of a Google loop. Explain machine learning to your grandmother, then
design an LLM-based Q&A system over company documents, read an eval sheet, and make a freshness call.
Written hiring report after.
| ||
46 | A practice room in the style of a Meta loop. Design a recommendation system, read an A/B sheet with a
guardrail that moved, then design a detector for policy-violating ads and pick its threshold. Written
hiring report after.
| ||
47 | A practice room in the style of an Amazon loop. Build a product with generative AI starting from the
customer, make a call on a pilot with a wrong-policy problem, then tell two stories about your own work
and be dug deep. Written hiring report after.
| ||
48 | A practice room in the style of an AI lab loop. Map and rank the risks of deploying an AI system, then
design a claims agent under a token budget, read a three-design pilot sheet and make the call. Written
hiring report after.
| ||
49 | ![]() | Three rounds of explaining a Gen AI concept in sixty seconds. A real question appears; you give
one plain line, the mechanism and the trade-off out loud. Ada scores it on screen, one line per skill,
and you retry the weakest. About six minutes.
| |
50 | ![]() | Three rounds of fixing a failing prompt. A failure appears; you name what is wrong, the missing
part of the prompt and one change, out loud. Ada scores it on screen, one line per skill, and you
retry the weakest. About six minutes.
| |
51 | ![]() | Three rounds of triaging an AI failure. An incident appears; you name the kind of failure, why
it happened and one defense, out loud. Ada scores it on screen, one line per skill, and you retry the
weakest. About six minutes.
| |
52 | ![]() | Three rounds of picking an eval. A real question appears; you say the measure, the test set and
the bar out loud. Ada scores it on screen, one line per skill, and you retry the weakest. About six
minutes.
| |
53 | ![]() | Three rounds of choosing prompt, RAG or tune. A situation appears; you pick, give the reason and
say how you would know, out loud. Ada scores it on screen, one line per skill, and you retry the
weakest. About six minutes.
| |
54 | The Full Gen AI Loop Document | ||
55 | A full Gen AI loop with Felix in one 45-minute call: four rounds with four hats, concept, build, risk and evals, and AI product. Written hiring report after.
| ||
56 | Take this at the end of the course. Twenty minutes with Felix, a parallel set of questions, the same five signals, and a comparison with your first report so you can see what moved.
| ||