AI Engineer course · Synchronous online
Build AI products that are worth shipping.
The engineer who builds AI into a product used by thousands, then tests it, launches it, measures it and keeps it fast and affordable. Every class is synchronous and hands-on.
- Weekend cohort
- From Sat, Jan 9
- Weekday cohort
- From Mon, Jan 11
- Length
- 6 weeks · 54 hours
- Price
- $3,900

01 · The role
What an AI engineer does.
An AI engineer is a software engineer on a product team who builds AI features that thousands of people use, then keeps them good, fast and affordable. The work is shipping, measuring and improving, not researching or training new models from scratch.
- Build streaming AI features and the screens around them
- Test AI code on every change and block quality drops before release
- Launch safely with feature flags, staged rollout and A/B tests
- Measure what users do and turn it into better data
- Fine-tune and run your own small models to cut cost and wait time
- Add voice, images and safety for a public app
02 · How it differs
Building with models, not training them.
vs. a data scientist
vs. a forward deployed engineer
Role titles vary from employer to employer. These descriptions summarize how the roles are commonly described.
03 · Why now
Demand for the AI engineer role is growing fast.
LinkedIn ranked AI engineer the fastest-growing US job title (Jobs on the Rise 2026, covering Jan 2023–Jul 2025).
AI engineer postings among the 639,000 AI-related US postings LinkedIn added between 2023 and 2025.
Share of US job postings mentioning AI skills in 2025, up 55% from 2024.
Sources: LinkedIn, Jobs on the Rise 2026 (Jan 7, 2026); CBS News reporting on LinkedIn data (Apr 16, 2026); Lightcast analysis in the Stanford AI Index Report 2026 (Apr 13, 2026). Job-title categories vary by source, and LinkedIn's "AI engineer" title can include roles also called machine learning engineers. These figures describe demand for a role. They are not a prediction of any individual's results, and AI FDE School does not guarantee employment, salary or any other outcome.
04 · Schedule
Two cohorts. Same course. Pick your week.
Saturday and Sunday
8 AM–12 PM ET
- First class
- Sat, Jan 9, 2027
- Last class
- Sat, Feb 13, 2027
- Hackathon
- Sun, Feb 21, 2027
Two sessions each day, so each weekend covers one full module.
Monday to Thursday
7–9 PM ET
- First class
- Mon, Jan 11, 2027
- Last class
- Tue, Feb 16, 2027
- Hackathon
- Sun, Feb 21, 2027
22 two-hour sessions. Classes run on Martin Luther King Jr. Day (Jan 18) and Presidents' Day (Feb 15).
05 · Curriculum
22 sessions. One product. Three releases.
Module 1Foundations, fastJan 9–10
- 01Sat Jan 9
How AI models work, and your first AI feature
How a model writes text one piece at a time; why the same question gets different answers; context limits; cost per call; big vs. small models
LabAdd a "Summarize this note" button to the study-notes app and record tokens, cost and wait time for every call
- 02Sat Jan 9
Prompts as code and structured output
Prompts kept in files and reviewed like code; asking for data in a fixed format; checking it and retrying when it is broken
LabGenerate flashcards from a note as structured data, check them and save them to the database
- 03Sun Jan 10
Tool calling and a small agent
Letting the model call your functions; the decide-act-check loop; when a fixed workflow beats an agent; asking the user before any change
LabA study assistant that searches the user's notes and builds a quiz, with a confirm step before it saves
- 04Sun Jan 10
Ask my notes: retrieval with citations, one user at a time
Embeddings and search; showing sources; keeping every user's data separate in a shared system
Lab"Ask my notes" with citations, plus an automated test proving one user can never see another user's notes
Module 2Build the AI feature users touchJan 16–17
- 05Sat Jan 16
The AI backend: streaming, timeouts and background jobs
Sending answers word by word; timeouts and retries; switching to a backup AI provider; queues for long jobs; limits on requests per user
LabA streaming endpoint with a backup provider, and a background queue for summarizing a 50-page PDF
- 06Sat Jan 16
Streaming chat and generative UI
Showing each AI action as a card on screen; partial results; stop and regenerate; progress indicators that make waits feel shorter
LabThe in-app assistant panel with streaming text, a "creating quiz" card, and stop and regenerate buttons
- 07Sun Jan 17
Designing for AI mistakes, and capturing feedback
Drafts users edit before accepting; undo; showing sources and uncertainty; error screens; thumbs up and down vs. silent signals (accepted, edited, hidden)
LabInline suggested flashcards with accept, edit and dismiss, each logged as an analytics event
- 08Sun Jan 17
Multi-user AI: quotas, metering and personalizationProject 1: launch v1
AI cost per user and per plan; free-tier limits; storing preferences safely; simple personalization; "delete my data"
LabA usage meter and free-tier quota, a "what to review next" suggestion and a working delete-my-data path
Module 3Quality you can shipJan 23–24
- 09Sat Jan 23
Testing AI code in everyday CI
Why AI code is hard to test; fake models and recorded answers; format checks; a small live smoke test on every code change; tests that pass only some of the time
LabA GitHub Actions test suite: fake-model unit tests, recorded-answer tests and a 20-case live smoke test
- 10Sat Jan 23
Traces, error analysis and evals
Recording every AI call; reading real conversations and labeling what went wrong; grouping failures; writing checks for the top failures; an AI judge checked against human labels
LabClassmates use each other's apps; label the recordings, build a failure list and write three evals
- 11Sun Jan 24
Prompt and model versions, release gates and upgrades
Versioned prompts with production and staging labels; one-click rollback; quality scores that block a bad change; handling a provider's new model
LabMove prompts into a prompt manager, ship v2 behind a label, fail the build when quality drops, then roll back
- 12Sun Jan 24
Product analytics for AI features
Defining success: acceptance rate, edits before accepting, return visits, cost per active user; event design; metrics that must not get worse
LabA dashboard for the app: suggestion acceptance, edit rate, slowest-request wait time and cost per active user
Module 4Launch and learn from usersJan 30–31
- 13Sat Jan 30
Launching safely: feature flags and staged rollout
Hiding new code behind an on/off switch; internal use first; releasing to 1%, 10%, 50% and 100% while watching metrics; a kill switch
LabRelease a new flashcard generator to 10% of simulated users, trigger an alert and use the kill switch
- 14Sat Jan 30
A/B testing prompts and models
Comparing two versions on real users; one main metric plus guardrail metrics; how many users you need; why peeking early misleads; a ship or no-ship memo
LabRun an A/B test of two prompt versions, compare acceptance and speed, and write a one-page decision memo
- 15Sun Jan 31
Feedback loops and data flywheelsProject 2: prove it and launch it
Turning user actions into training and test data; small curated data beats large messy data; consent before training on user data
LabTurn logged accept, edit and reject events into a clean, consent-checked dataset of about 500 examples
Module 5Own the model: custom, small, fast and cheapJan 31–Feb 7
- 16Sun Jan 31
Fine-tuning a small model
When to fine-tune (format and behavior) and when not to (facts belong in retrieval); LoRA in plain terms; the full cost of owning a model
LabFine-tune a small open model on your own data on a GPU we provide, and compare it with a big model on your evals
- 17Sat Feb 6
Distillation: a small model that learns from a big one
Teacher and student models; using a big model to write and grade examples; filtering bad examples; comparing quality, speed and cost
LabGenerate and filter about 1,000 examples, train the small model and fill in a quality, speed and cost table
- 18Sat Feb 6
Running your own model
Open models; serving engines such as vLLM; quantization (smaller, faster models); keeping the same API; self-hosting vs. pay-per-call math
LabServe your model, point the app at it, benchmark full vs. quantized versions and work out cost per 1,000 requests
- 19Sun Feb 7
Speed and cost at scale
Prompt caching; exact and similarity caches; routing easy requests to the small model; gateways with backups; parallel steps; batch jobs
LabAdd caching and a router between your small and big models, then report before-and-after wait time and cost per user
Module 6Rich features and trust at consumer scaleFeb 7–13
- 20Sun Feb 7
Voice features in real time
Speech-to-speech vs. speech-to-text, model, text-to-speech; turn-taking and interruptions; wait-time targets; tool calls by voice; cost per minute
LabA "quiz me out loud" voice mode, with a tool call that saves the score
- 21Sat Feb 13
Images, documents and on-device AI
Photos and PDFs as input; turning handwritten notes into structured notes; when a small on-device model is enough and when to use the cloud
Lab"Snap your notes": photo to structured note to flashcards, plus an offline fallback design
- 22Sat Feb 13
Trust at consumer scale and launch reviewProject 3: fast, cheap and rich
Layered safety checks; hiding personal data; hidden instructions planted in shared content; free-tier abuse; blocked attacks vs. false alarms; a launch checklist
LabAdd a safety layer, run an attack test set, then a mock "design an AI feature for a million users" interview
You build one multi-user AI study-notes app across the course and ship it in three releases. Each class is two hours: teaching, a guided lab and interview practice. Topics and order may be adjusted before the cohort starts.
How classes run
Synchronous classes. Never prerecorded.
Most online courses are self-paced, prerecorded videos. Ours are synchronous: you, your classmates and your instructor log in at the same scheduled time, cameras on, and learn together like an in-person class.
Same time, every class
Everyone logs in together at a scheduled time: weekday evenings or weekend mornings, Eastern time.
Cameras on, like a classroom
You see your instructor and classmates, and they see you. Need a short break from the camera? Tell your instructor before class.
Instructor-led and hands-on
Your instructor teaches, then you build in class while they watch, answer and fix problems with you.
Attendance counts
Attend at least 18 of the 22 sessions for your certificate. Classes are not recorded, just like a university classroom. Missed a class? Join the same session in the other cohort if it is still ahead, or catch up with that session's lab code and notes and bring your questions to office hours. Miss more than 4 classes and you can move once, free, to the next cohort of the same course.
| Self-paced video course (asynchronous) | Webinar | AI FDE School: synchronous | |
|---|---|---|---|
| When you learn | Whenever you like | At a set time | At a set time, with the same classmates all course |
| Cameras | None | Usually only the presenter | Everyone, cameras on |
| Hands-on work | Alone, after the video | Rarely | In every class, with your instructor watching |
| Questions | Forum, answered later | Chat box, if time allows | Ask out loud, answered on the spot |
| Attendance | Not tracked | Not tracked | Expected: 18 of 22 sessions for the certificate |
| Recordings | The videos are the course | Often recorded | Not recorded, like a university classroom |
Included with every seat
More than the synchronous classes.
Live office hours + 1-on-1s
Live group office hours twice a week, plus two 20-minute 1-on-1 sessions with your instructor per course.
24/7 help channel
Ask anytime, 24/7. Your instructor replies within 24 hours.
LCAE exam prep, fee back on a pass
Prep for the LangChain Certified Agent Engineer (LCAE) exam: LangChain, LangGraph and LangSmith. Pass within 3 months and we reimburse the exam fee. Claude certification prep too: if your employer is a Claude Partner Network member, you can take Anthropic's Claude exams, and we reimburse those on a pass as well.
Certificate of course completion
Attend at least 18 of the 22 synchronous sessions and submit the projects.
06 · Price
Synchronous online classes, never prerecorded. US dollars. One price for the full course, hackathon included. Or 3 monthly payments of $1,350 ($4,050 in total), and employers can pay by invoice.
- 22 synchronous, instructor-led sessions (44 hours), cameras on, never prerecorded
- 10-hour AI Engineer Hackathon with prizes
- Your own AI app, shipped in three releases
- Lab computing included: GPU time for fine-tuning and running models, model hosting and voice usage
- Live group office hours twice a week
- Two 20-minute 1-on-1s with your instructor
- 24/7 help channel (instructor replies within 24 hours)
- LangChain LCAE certification prep; exam fee reimbursed when you pass (Claude exams too, if your employer is a Claude partner)
- Certificate of course completion
- Weekend or weekday cohort
- Full refund within your first 16 class hours: until Sun, Jan 17 (weekend) or Thu, Jan 21 (weekday) (Refund Policy)
AI Engineer Hackathon
Launch Day. Real users. $10,000 in prizes.
Sunday, February 21, 2027, 8 AM–6 PM Eastern. Teams build and launch an AI feature, use each other's apps, and are judged on measured quality, speed, cost, safety and real usage.
Open to enrolled AI Engineer students, alone or in teams of up to four. A team's prize is split equally. Official rules will be published before the event.
07 · Details
Course details.
- Format
- Synchronous online classes: hands-on, taught in real time in a virtual classroom, with Q&A.
- Type
- Non-credit skills training. Not a degree; no academic credit. Certificate of course completion when you attend 18 of the 22 sessions and submit the projects.
- Support
- Live group office hours twice a week, two 20-minute 1-on-1s, a 24/7 help channel (instructor replies within 24 hours), and prep for the LangChain Certified Agent Engineer (LCAE) exam, with the exam fee reimbursed when you pass within 3 months (Claude exams too, if your employer is a Claude partner).
- Length
- 22 two-hour sessions (44 hours) plus the 10-hour hackathon, 54 hours in total.
- Cohorts
- Weekend: Saturday and Sunday, 8 AM–12 PM ET, from Jan 9. Weekday: Monday to Thursday, 7–9 PM ET, from Jan 11. See the schedule.
- Before you start
- You should be comfortable writing code and building a simple web backend. No AI experience is needed.
- Lab costs
- Included. We provide the GPU time, model hosting and voice usage the labs need.
- Seats
- Limited, to keep each group small and focused.
- Taught by
- Trainers with 25+ years of industry experience, including work at startups, Oracle, Amazon, Google and Meta. Past experience only; AI FDE School is independent and not affiliated with or endorsed by these companies.

Who it's for
- Backend, full-stack and product engineers who want to ship AI features
- Engineers on product teams adding AI to an existing app
- Technical founders building an AI product
- FDE course graduates: about 16 of the 22 sessions are new material
08 · Questions
About the AI Engineer course.
When does the course start?
How much does it cost?
How many hours is the course?
What is an AI Engineer?
I'm taking the FDE course. Is this course different?
Are lab costs extra?
How is this different from machine learning engineering?
Do I need AI experience already?
Are the classes synchronous or self-paced?
Is this a degree or certificate program?
Are classes synchronous, or self-paced videos?
Is there help between classes?
Do you help with industry certifications?
Can I join from outside the USA?
Seats are limited. Start Jan 9 or Jan 11.
Join a free, live info session on a Saturday morning to see if AI Engineer fits you, and ask the trainers anything.
