Mark My Words is a rapid-growing EdTech platform designed to transform how English teachers assess student work, saving them time and enhancing feedback quality. Our AI-powered tools transcribe handwritten essays, track student progress against curriculum skills, and help educators deliver timely, personalised, and impactful feedback. We are a passionate team dedicated to reducing teacher workload and improving student outcomes.

About the role

We're looking for a Machine Learning Engineer to own and continuously advance the OCR and document-understanding stack that turns messy, real-world images of student handwriting into clean, structured text. You'll work mostly on vision and OCR problems—primary-school handwriting, varied lighting and camera angles, scans, and PDFs—and you'll be responsible for the full lifecycle: shipping models, making them smaller and faster, building the evaluation infrastructure that proves they're improving, and directing the labelling team that feeds them.

What you'll do

- Build, fine-tune, and deploy OCR and handwriting-recognition models for noisy, real-world student work across a wide range of ages and writing styles.
- Work with vision-language models (VLMs) for end-to-end document understanding—reading handwriting, preserving layout and structure, and handling edge cases that traditional OCR pipelines miss.
- Fine-tune and align models using techniques such as supervised fine-tuning and reinforcement learning (e.g. TRL), and run experiments to push recognition accuracy on our hardest cases.
- Continuously improve the model. Run a steady cadence of experiments, ship incremental gains, and close the loop between real production usage and the next round of training.
- Make models smaller, faster, and cheaper. Apply quantisation, distillation, pruning,



and inference optimisation to cut latency and cost while holding or improving accuracy—so we can process entire class sets in minutes at scale.
- Design and own evaluation. Define the metrics that matter (character/word error rate, normalised edit distance, layout and structure fidelity, downstream feedback quality), build robust eval sets that reflect the diversity of real student writing, and create the harnesses and dashboards that make model quality measurable and regression-proof.
- Lead the labelling team. Own data quality as the primary driver of model improvement: set annotation guidelines, define what gets labelled and why, manage throughput and quality, and build the tooling and feedback loops (including active learning) that turn labelling effort into measurable model gains.
- Collaborate with the broader engineering team to take models from prototype to reliable production services.
- Monitor models in production, diagnose failure modes, and feed what you learn back into both training and labelling priorities.

What we're looking for

- Strong applied machine learning experience, with a focus on computer vision and/or OCR/document AI.
- Solid understanding of vision-language models and how to use, fine-tune, and evaluate them.




- Hands-on experience with OCR—either training/fine-tuning recognition models or building production OCR pipelines—ideally including handwriting.
- Experience fine-tuning and aligning models, including familiarity with libraries such as Hugging Face TRL and the broader Transformers/PyTorch ecosystem.
- Experience making models efficient for production—quantisation, distillation, pruning, or inference optimisation—and reasoning about accuracy/latency/cost trade-offs.
- A genuine instinct for evaluation: you believe you can't improve what you can't measure, and you've built metrics and eval pipelines before.
- Experience directing data annotation or labelling efforts, and a clear view of how data quality translates into model quality.
- Strong Python and software engineering fundamentals, with experience shipping ML to production.
- Comfort working with ambiguous, messy real-world data rather than clean benchmarks.

Nice to have

- Experience with handwriting recognition specifically, or with low-resource / domain-specific OCR.
- Experience building data annotation tooling, active-learning loops, or managing a labelling team.
- Familiarity with model serving and inference optimisation at scale (e.g. vLLM, or similar).
- An interest in education and the impact of getting this right for students and teachers.

Why join us

- Your work has direct, visible impact—better OCR means better feedback for real students and hours saved for real teachers.
- A meaty, well-defined technical problem at the core of the product, with full ownership of the model and the data pipeline behind it.

Be careful - Don't provide your bank or credit card details when applying for jobs. Don't transfer any money or complete suspicious online surveys. If you see something suspicious, report this job ad.

#J-18808-Ljbffr

📌 Machine Learning Engineer (Melbourne)
🏢 Mark My Words
📍 Melbourne