CV
Computer Vision
Detection, segmentation and analysis for challenging imagery — including satellite and remote-sensing pipelines where labels are scarce and quality matters.
// freelance ML engineering
Practical machine learning for vision, multimodal & data-centric AI.
MaineFrame Labs is a one-person ML engineering studio led by a PhD researcher with industry experience — building production-grade computer vision and multimodal generative AI systems, with a data-centric edge grounded in active learning and uncertainty quantification.
Focus areas
What I work on
Three tightly related disciplines, brought together under one roof so models ship reliable and stay reliable.
MM
Multimodal Generative AI
Building with vision-language models and multimodal pipelines — and critically, understanding their limits through targeted evaluation and explainability work.
DC
Data-Centric AI
Active learning loops and uncertainty quantification to spend labeling budget where it matters and to know when a model's predictions can be trusted.
End to end
I own the entire pipeline
From the first look at the raw data to monitoring a model in production — one accountable engineer (me) across every stage, so nothing falls through the gaps between hand-offs.
01
Data inspection & cleaning
Profiling, labeling strategy, and cleaning — catching the label noise, leakage, and distribution oddities that decide whether a model can succeed at all.
02
Modeling & evaluation
Training with an honest evaluation harness, calibrated uncertainty, and active learning — measuring what matters, not what's easy to metric.
03
Deployment
Reproducible pipelines and packaging so the model that shipped is the model that was evaluated — deployed into your environment, not a notebook.
04
Post-deployment monitoring
Drift, calibration, and performance tracking after launch, with feedback loops that feed straight back into data and the next training cycle.
Engagement
Have a problem that needs a model?
Freelance, project-based, or embedded with your team. If it touches vision, multimodal AI, or data quality, it's probably a fit.