Research Scientist, Foundation Model Training, SeekrGEO

Seekr • Austin, TX

1d • Hybrid

About The Position

SeekrGEO is Seekr’s geospatial AI product. This role contributes to the foundation model program behind it: pretraining and post-training of large multi-modal models on geospatial data, together with the distributed training systems that make that work possible at scale. The focus is training, but the role supports the full model lifecycle through deployment. As a Research Scientist you lead the modeling side of that program: architecture choices, data mixture, pretraining and post-training recipes, and the evaluations that tell us whether the model works. You will work alongside Research Engineers on the systems that make these workloads run at scale, and you prototype recipes in code rather than only on paper.

Requirements

• Strong background in modern ML, with deep familiarity with transformer architectures, multi-modal models, and the practical realities of training them at scale.
• Demonstrated experience taking a large model through pretraining or continued pretraining, not just fine-tuning a frontier checkpoint.
• Experience designing and running post-training pipelines (SFT, preference optimization, RL) and reasoning clearly about their failure modes.
• Comfort designing experiments, evaluating ambiguous technical tradeoffs, and building evals that hold up under scrutiny.
• Ability to move comfortably between research and engineering. You write production-grade training code, not only experiment scripts, and you can collaborate credibly with engineers on scaling decisions.
• Expert-level PyTorch fluency, with a deep understanding of distributed training strategies (FSDP, tensor / pipeline / sequence parallelism, ZeRO, checkpointing) and how each interacts with optimization dynamics, effective batch size, and model behavior.
• Strong Python and software engineering fundamentals, with comfort in testing, code review, debugging, and iteration under production pressure.
• Fluency with AI coding assistants and the modern developer workflows they enable.
• Clear communication and strong collaboration across technical and non-technical partners.
• Reside near Austin, TX or Reston, VA and able to work 3 days per week in office.

Nice To Haves

• Experience with remote sensing modalities (SAR, hyperspectral, multispectral, high-cadence EO) and the modeling quirks of each.
• Experience training or post-training agentic systems, tool-using models, or planner / executor architectures.
• Familiarity with government and defense data handling, classification regimes, or air-gapped deployment.
• Experience with model distillation, quantization, or other techniques for meeting real latency and cost constraints in deployed systems.
• Open-source contributions to ML research code or geospatial ML libraries.

Responsibilities

• Own the pretraining and continued pretraining strategy for our multi-modal geospatial foundation models: model architecture choices, data mixture, objective design, and the ablations that justify the choice.
• Design and run post-training pipelines spanning SFT, preference optimization, RL, and reasoning-focused recipes for multi-modal models.
• Design evaluations that are meaningful, reproducible, and trusted both by us and by customers. Build the benchmarks and probes that surface real capabilities and real failures.
• Diagnose training-dynamics problems (loss spikes, gradient pathologies, optimizer instabilities, capacity issues, data mixture drift) and partner with Research Engineers to turn fixes into durable platform improvements.
• Curate and version training corpora with the rigor required for repeatable runs and meaningful ablations. Decide what data belongs in each mixture and why.
• Read current literature and translate promising ideas into shipped capabilities. Reproduce a paper’s core claim, stress-test it on our workloads, and decide whether it earns a place in the next training cycle.
• Support deployed models through their lifecycle: monitor behavior in production, diagnose regressions in the field, and close the loop back into the next training cycle.
• Partner with the SeekrGEO product team and customer-facing teams to align training objectives with the workflows the model needs to support.
• Write production-grade training code alongside Research Engineers, staying hands-on through the training and evaluation loop.
• Use AI coding assistants effectively as part of a modern research workflow while maintaining strong judgment over modeling choices, training code, and evaluation logic.

Benefits

• Equity Ownership – RSUs that let you share directly in Seekr’s long‑term success and growth.
• Time Off That Respects Real Life – Unlimited PTO plus 14 paid company holidays to truly recharge.
• Work Your Way – A flexible hybrid work environment with offices in Reston, VA and Austin, TX.
• Competitive Total Rewards – A role‑appropriate compensation structure that supports long‑term growth, including base salary, bonuses, or commission plans depending on role.
• 401(k) with Company Match – Build your future with a retirement plan that includes employer matching.
• Comprehensive Health & Wellness – Medical, dental, vision, and life insurance coverage starting day one—for you and your family.
• Parental Leave – Paid parental leave to support employees as they welcome a new child through birth, adoption, or foster placement.

What This Job Offers

• Job Type: Full-time
• Career Level: Senior