Own the allocation solver beyond the CP-SAT baseline. Role shape: Own and advance Kyber's allocation solver beyond the CP-SAT baseline: more constraints, faster and more stable re-solves, and objectives informed by learned models rather than hand-encoded rules. Environment: Python, CP-SAT and OR-Tools, MIP and metaheuristics, incremental re-optimisation, benchmarking on live station data, Postgres, AWS, close work with the machine learning role. Experience: Minimum three years in applied optimisation or operations research. Practical reality: direct customer contact, time in live operations starting on the airport ramp, coding credits, London-based, remote only by exception, UK work authorization preferred. The Problem: Operations in critical infrastructure, in transportation and logistics, look organized from the outside, but behind the curtain they run on radios, Excel sheets, emails, paper manifests, phone calls, and constant improvisation. Plans are built statically ahead of time for an operation that changes by the minute: delays, location changes, sick calls, broken equipment, and knock-on effects across the system. Humans cannot manage that complexity at operating speed, so teams compensate by overstaffing and building slack into the operation - and still end up missing people in the moments that matter most and cost the most. What Kyber Brings: Kyber builds the plan before operations start, then turns it into live control as the day unfolds. We use algorithms, stateful systems, and agentic AI to continuously recompute the best allocation of tasks, staff, equipment, locations, and time as conditions change - something humans cannot execute reliably at operating speed. In aviation, that means replacing spreadsheets, radios, fixed teams, and human heroics with real-time coordination at the turnaround. The old playbook no longer holds, and the technology is finally good enough to replace it. The Challenge: Assigning workers to tasks under hard constraints is NP-hard and has to run at operational speed. The baseline is CP-SAT with lexicographic objectives, and it solves a deliberately simplified version of the problem. Making it realistic means widening the model along several axes at once: the spatial cost of moving between locations, conditions that change how long work takes, worker skill and eligibility, working-time and rest rules, and the preferences operators hold but rarely state. Each axis adds variables and couples constraints that were previously independent, so the search space grows faster than the model gains accuracy. Two properties pull against that. Re-solve on disruption has to complete in seconds, and successive plans have to be stable, because churn costs more than the objective gain it buys. Constraint sets also differ by site and by industry, so they have to be parameterised rather than rewritten, and the parts of the objective that encode human judgement have to be learned rather than enumerated. That puts the interface with the machine learning work inside the solver design, and it makes eliciting the constraints nobody writes down part of the role. What We Are Looking For: You own the solver: model formulation and decomposition; warm starts and incremental re-solve; heuristics and metaheuristics where exact methods stop scaling; objective and stability design; a benchmarking harness built on real operational data. You decide what stays exact, what becomes learned, and where the boundary sits. The measure is a plan an operations manager trusts enough to run their shift on, and that holds when the day goes wrong. That means going on site, understanding why a plan was rejected, and turning it into a constraint or a cost. At least three years in applied optimisation, with production systems behind you. Depth in constraint programming with CP-SAT or OR-Tools, mixed integer programming, or metaheuristics such as large neighbourhood search, applied to scheduling, rostering, or routing. Strong Python. Real-time and incremental re-optimisation, column generation, and learning-augmented optimisation all count, as does experience making a solver stable and explainable enough for operators to trust it. The bar is high because the problem is hard, and Kyber will only become an exceptional company if the people building it are exceptional. We look for highly agentic engineers who move without being told what to do next, take on hard problems, and hold their standards when the operation is live and the stakes are real. Salary is not a constraint for the right person, and the role comes with coding credits. London-based, remote only by exception. UK work authorization preferred.