The posting
About Applied Computing
Applied Computing was founded in 2024 to build Orbital, a physics-informed foundation model for energy operations. We’re live across oil and gas, refineries, and petrochemicals, working towards our mission: sustainable abundance for a growing planet.
The hydrocarbon industry keeps the world running. But its complexity has left operators tied to legacy systems, making critical decisions on less than 10% of available data.
We built Orbital to change that. It’s a foundation model built specifically for energy that lets companies use AI at scale, harnessing all of their operational data and optimising in real time for any metric. Decisions get faster, operations get safer, and carbon intensity falls.
We’ve raised over $32 million, including one of the largest seed rounds for an AI company in the UK. We’re just getting started.
The Role
The Time Series Researcher owns the core of Orbital’s temporal intelligence. This role exists to design, validate, and deploy foundational time-series models that operate under real world constraints: noisy sensors, partial observability, physical laws, and high economic stakes.
This is a customer facing role part of forward deployed team working with customer engineering teams, fine tuning our model to their plants.
This is not offline research. You will own the full lifecycle; from theoretical formulation and experimentation to real-time inference, uncertainty estimation, and continuous retraining in production.
What You’ll Own • Orbital’s foundational time-series modelling stack • Physics-informed and probabilistic model design • Uncertainty quantification and robustness under sensor faults • Research → production translation for time-series models • Benchmarking standards and validation protocols used across the company
Must-Have Qualifications
- PhD in Computer Science, Statistics, Applied Mathematics, Physics, or related field
- First-author publications in time-series modelling, forecasting, signal processing, or physics-informed ML
- 3+ years of hands-on research experience in time-series or sequence modelling
- Demonstrated experience in Deep Learning & Probabilistic modelling
- Expert Python skills with production-grade PyTorch code
- Experience deploying ML models into real systems
How We Work • Research is judged by production impact, not paper count • We value principled models that survive contact with reality • We iterate aggressively, benchmark honestly, and ship responsibly • Physics, statistics, and learning are treated as complementary, not competing
What This Role Is Not • Not offline academic research disconnected from deployment • Not pure deep-learning experimentation without domain grounding • Not feature engineering on static datasets • Not a support role; this position owns core IP
1. Design & Implement Foundational Time Series Models • Design core time-series architectures supporting: o Forecasting o Classification / anomaly detection o Optimisation & control-adjacent tasks • Explore and select appropriate objectives examples: o Probabilistic losses o Generative formulations o Reinforcement-learning-inspired objectives where appropriate • Develop hybrid approaches that blend: o Classical statistical models o Deep learning architectures o Physics-based constraints
2. Embed Physics-Informed Structure
- Integrate domain physics into learning systems, including: o Conservation laws o Process constraints o Differential-equation-based priors
- Improve generalisation, interpretability, and extrapolation beyond training regimes
- Ensure models respect physical feasibility in production settings
- 3. Uncertainty, Robustness & Reliability • Design uncertainty-aware models (Bayesian, ensemble, hybrid) • Quantify confidence under: o Sensor drift and failure o Regime change o Sparse or delayed ground truth • Ensure outputs are usable by operations and engineering teams, not just statistically elegant. 4. Production Structured AI Code • Containerise and deploy models using Docker on AWS / Azure (EKS, ECS, SageMaker) • Build or integrate CI/CD pipelines for: o Training o Evaluation o Rollout and rollback o Automated retraining triggers 5. Benchmarking & Validation • Define rigorous back-testing and evaluation protocols • Build automated benchmarking pipelines across datasets, regimes, and failure modes • Compare against classical baselines and modern deep-learning approaches • Ensure claims are defensible to customers, partners, and internal stakeholders What Success Looks Like First 90 Days • Deep understanding of Orbital’s data, domains, and production constraints • Contribution to at least one core time-series model or experimental track • Clear ownership of a modelling problem with defined success metrics 6–12 Months • One or more foundational models running reliably in production • Demonstrable improvements in: o Forecast accuracy o Robustness under faults o Uncertainty calibration • Models actively used by downstream agents and optimisation layers • Benchmarking standards adopted across the research team
