About Flashmind Labs
We are building foundation models that engage with reality directly and evolve beyond their training data. Models that are calibrated, unconstrained in their pursuit of truth, and built for decisions with real consequences.
We are a focused team of researchers and engineers in Dubai. Come build the future of learning systems with us.
About this Role
We're looking for researchers to push the boundaries of what foundation models can do when optimized for truth. You will work on novel training objectives, architectures designed to preserve calibration and handle uncertainty natively, and evaluation frameworks grounded in empirical outcomes.
You will work with a team of scientists and engineers on core research efforts, including:
- Novel training objectives for factual accuracy and calibration
- Architectures that handle uncertainty and distributional information natively
- Training pipelines grounded in empirical outcomes
- Evaluation frameworks for measuring calibration, tail event accuracy, and robustness under distribution shift
Minimum Qualifications:
- PhD or equivalent experience in ML, CS, or a related field
- Proficiency in Python and PyTorch
- Ability to design, run, and analyze experiments independently
- Understanding of large-scale training and accelerator-based compute environments
Preferred Qualifications:
- Deep expertise in at least one of: alignment methods (RLHF, DPO), calibration, loss function design, or model architectures
- Demonstrated record of contributing to advanced research via publications and/or major model releases
- Experience with distributed training at scale
- Background in epistemology, calibration, or decision theory