From Teacher to ML Engineer: A Real AI Career Change Success Story
Category: Case Study --- Introduction: The Hook It's 9:47 PM on a Tuesday, and Maya is staring at a wall of logs from a failing training run.
Category: Case Study
Introduction: The Hook
It's 9:47 PM on a Tuesday, and Maya is staring at a wall of logs from a failing training run. Her model's loss curve has flatlined, and somewhere in a Docker container, a dependency is quietly breaking her pipeline. She finds the bug — a mismatched tokenizer version — fixes it, and reruns the job. At 10:30 PM, the validation accuracy ticks up. She smiles, closes her laptop, and goes to bed.
Eighteen months earlier, that same evening would have looked very different. Maya was a high school math teacher grading a stack of 140 quizzes, wondering if there was another career where her analytical brain could actually stretch.
Today, she's a Machine Learning Engineer at a mid-sized edtech company, earning $142,000 — up from a teacher's salary of about $58,000.
Maya's story matters because it busts one of the most persistent myths in tech: that AI careers are reserved for computer science graduates who've been coding since they were twelve. They're not. AI teams need people who understand problems, communicate clearly, and can learn fast — and teachers happen to be exceptional at all three.
In this case study, you'll follow Maya's full 18-month journey: how she chose her target role, what she learned (and in what order), the four projects that got her interviews, and how a single referral changed everything. If you're a teacher, a career-changer, or anyone wondering whether you can break into AI, this roadmap is for you.
Before AI: Maya's Starting Point
Maya was 32 when she started. She had a bachelor's degree in mathematics, seven years of high school teaching experience, and exactly zero lines of production code to her name.
Her transferable skills were stronger than she realized:
- Mathematics: Linear algebra and statistics weren't foreign concepts — she'd taught them.
- Communication: Explaining backpropagation to a hiring manager is, in many ways, just lesson planning.
- Curriculum design: Structuring a learning path is a core skill for self-taught engineers.
- Patience with ambiguity: Teachers debug human confusion daily.
The turning point came in early 2023. A colleague showed her ChatGPT, and Maya spent a weekend experimenting. She tried building a tool that would auto-generate practice problems aligned to her curriculum. It was clunky, but it worked — and she was hooked. Combined with growing burnout from 60-hour teaching weeks, the pull toward tech became impossible to ignore.
Her honest starting position was humbling: no coding experience, no tech network, and a tight budget (she could afford maybe $500 total for courses). Her fears were the classic ones: Am I too old? Do I need a CS degree? Will anyone take a teacher seriously?
Lesson Learned: Your existing domain expertise is an asset, not a liability. Maya's math background and teaching skills became differentiators in interviews — she could explain models clearly and connect ML to real educational problems. Don't erase your past career; leverage it.
Choosing a Target Role
The AI job market is broad, and Maya spent two weeks mapping it before committing. Here's the landscape she evaluated:
| Role | Typical Salary (US) | Entry Barrier |
|---|---|---|
| ML Engineer | $120K–$250K | High (coding + ML depth) |
| NLP Engineer | $130K–$220K | High |
| Computer Vision Engineer | $130K–$230K | High |
| Prompt Engineer | $80K–$180K | Medium |
| AI Product Manager | $130K–$220K | Medium (needs PM experience) |
| Data Scientist | $100K–$180K | Medium-High |
| AI Solutions Architect | $150K–$250K | High (usually senior) |
She evaluated each on three axes: salary ceiling, entry barriers, and skill fit. AI PM roles required product experience she didn't have. Data Science felt crowded and less engineering-focused. Prompt Engineering was interesting but had a lower salary ceiling and fewer stable roles.
She chose ML Engineer because it combined her math strength with a clear, learnable skill stack — and it paid well. Prompt Engineering became a stepping stone project area, not a destination: building a prompt-evaluation case study let her demonstrate modern LLM skills early while she built deeper engineering chops.
The reality check came from reading 40 real job descriptions. The pattern was consistent: Python, PyTorch or TensorFlow, SQL, cloud (AWS/GCP), Docker, and Git. Some listed a CS degree "or equivalent experience." That "or" mattered.
Lesson Learned: Pick one target role early; explore others only through projects. Spreading yourself across five roles is the fastest way to master none.
The Learning Journey: Skills and Tools
Maya structured her learning into four phases over roughly 14 months.
Phase 1 — Foundations (Months 1–3)
She started with Python using freeCodeCamp and Harvard's CS50P. She learned syntax, object-oriented programming, virtual environments (venv), and Git/GitHub. In parallel, she refreshed linear algebra (vectors, matrices, dot products), statistics, and calculus essentials via Khan Academy and 3Blue1Brown.
# Maya's first "real" script: a grade calculator
def weighted_average(scores, weights):
return sum(s * w for s, w in zip(scores, weights)) / sum(weights)
print(weighted_average([88, 92, 79], [0.3, 0.5, 0.2])) # 87.1
Phase 2 — Core ML (Months 4–7)
She moved to NumPy, pandas, and scikit-learn, working through Andrew Ng's Machine Learning Specialization on Coursera and fast.ai's Practical Deep Learning course. She learned supervised learning fundamentals: regression, classification, cross-validation, and model evaluation.
Phase 3 — Deep Learning and Modern AI (Months 8–11)
This was the deep end. She learned PyTorch basics — tensors, autograd, training loops — then neural networks, CNNs, and transformers. She explored the Hugging Face ecosystem, fine-tuning small models like DistilBERT and experimenting with the transformers library.
She also used ChatGPT and Claude as learning accelerators — asking them to explain concepts, review her code, and generate practice problems. But she learned their limits fast: they confidently produced wrong answers, so she always verified against documentation.
from transformers import pipeline
classifier = pipeline("sentiment-analysis")
print(classifier("This feedback was actually helpful."))
# [{'label': 'POSITIVE', 'score': 0.999}]
Phase 4 — Production and Engineering Skills (Months 12–14)
This phase separates hobbyists from hireable engineers. Maya learned SQL, Docker, cloud basics on AWS (S3, EC2, SageMaker), building APIs with FastAPI, and core MLOps concepts: model versioning, CI/CD, and monitoring.
Her biggest challenges were the math gaps (fixed with targeted review, not a full degree), tutorial hell (broken by forcing herself to build without tutorials), and time management — she studied 12–15 hours per week around a full-time job, mostly early mornings and weekends.
Lesson Learned: Build depth in Python + PyTorch + one cloud platform; breadth can wait. Recruiters want proof you can go deep, not a shallow tour of ten tools.
Projects: The Portfolio That Got Interviews
Maya built four projects, each solving a problem she genuinely cared about. This authenticity showed in interviews.
Project 1 — Classroom Analytics Tool. Using pandas + scikit-learn, she built a tool that predicted which students were at risk of falling behind based on quiz and homework patterns. It was her bridge from teaching to ML.
Project 2 — Fine-Tuned NLP Model for Essay Feedback. She fine-tuned a Hugging Face model to give automated feedback on student essays, training on a small labeled dataset she created herself. This demonstrated the full ML lifecycle: data collection, training, evaluation.
Project 3 — RAG Chatbot Web App. She built a retrieval-augmented generation chatbot over a curriculum knowledge base using FastAPI + Docker + an LLM API, deployed it as a working web app. This was her most impressive project — it showed she could ship.
Project 4 — Prompt Engineering Case Study. She documented a systematic evaluation of prompt variants for generating math problems, complete with metrics and iteration logs.
She documented everything meticulously: clean GitHub READMEs, a simple portfolio site, and short write-ups on LinkedIn explaining what she built and why.
Lesson Learned: Three polished, deployed projects beat ten notebooks. Every project should be runnable by a stranger in under five minutes.
Networking and Breaking In
Maya treated networking like teaching in reverse: ask good questions and share what you learn.
She optimized her LinkedIn profile, posted weekly project updates, and joined Kaggle competitions and Discord/Slack communities for ML learners. She attended local meetups and virtual conferences, where her communication skills — honed in classrooms — made her stand out in rooms full of nervous engineers.
She did cold outreach and informational interviews, reaching out to ML engineers at edtech companies with specific, thoughtful questions. Most ignored her; a few replied. One of those conversations led to a referral that produced her first interview.
Her application strategy was targeted, not scattershot: about 60 applications over three months, heavily focused on edtech and mission-driven companies where her background was an advantage. She was rejected 55 times. She kept a rejection log and treated each as data.
Lesson Learned: Networking is teaching in reverse — ask good questions, share what you learn. A referral beats a hundred cold applications.
Timeline and Milestones
Here's Maya's 18-month roadmap:
- Months 1–3: Python + math foundations (CS50P, Khan Academy)
- Months 4–7: Core ML coursework + Project 1
- Months 8–11: Deep learning, PyTorch, Projects 2–3
- Months 12–14: Production skills (Docker, AWS, FastAPI) + Project 4
- Months 15–16: Networking, portfolio polish, informational interviews
- Months 17–18: Applications, interviews, and offer
She landed the ML Engineer role at month 18, with a $142,000 base salary plus equity.
Conclusion: Your Roadmap Starts Today
Maya's story isn't magic — it's a repeatable pattern. She chose one target role, built deep skills in Python, PyTorch, and AWS, shipped four real projects, and networked with curiosity instead of desperation. Her teaching background wasn't a handicap; it was her edge.
If you're standing where Maya stood, here's your actionable starting point:
- Pick your target role this week. ML Engineer, NLP Engineer, or Data Scientist are the most teacher-friendly entry points.
- Start with Python and Git — freeCodeCamp and CS50P cost nothing.
- Commit to 10–15 hours per week and protect that time.
- Build your first project around a problem you already understand.
- Show up in communities and start conversations before you need a job.
The AI industry needs people who can think clearly, explain ideas, and solve real problems. Teachers do that every single day. The question isn't whether you can make the leap — it's whether you'll start.
Ready to map your own AI career path? Explore more role guides, salary data, and learning roadmaps at AICareerFinder.
🎯 Discover Your Ideal AI Career
Take our free 15-minute assessment to find the AI career that matches your skills, interests, and goals.