About Pelico

We are creating the factory of the future where disruptions are avoided, processes synchronized and value captured. Our factory operations management platform enables factory teams to be more agile and resilient in a world where supply chain disruptions occur every 16 minutes. Pelico empowers users to identify bottlenecks, avoid problems, and focus on innovation instead of fire-fighting. Since our foundation in 2019, we’ve partnered with industry leaders across aerospace, industrial equipment, and luxury watchmaking, revolutionizing factory operations in over 15 countries. Esteemed clients include Airbus, Safran, Cartier, Daikin, and Eaton.

Our Team

With a dynamic team of over 180 professionals across the US and France, Pelico is a melting pot of top-tier talent from Tech, Data Science, and Manufacturing domains.
Our collaborative environment fosters innovation and excellence, driving us to solve complex challenges and shape the future of manufacturing.

Our work has been recognized by Safran (Digital Transformation Award) and Microsoft (scale-up of the year).

Engineering turns Product ambition into scalable, reliable solutions that empower on-site operational teams.

We design a platform that handles high volumes, concurrent users, and complex workflows, while remaining modular, maintainable, and future-proof.

About the role

Pelico is the supply-chain digital twin for complex discrete manufacturers in aerospace, defense, and industrial equipment. We ingest the ERP and operational data, build an allocation graph of supply, stock, and demand, and propagate time through it to project — realistically — what the plant will actually be able to produce and ship, and where it will fall short. Our customers' central pain is missing parts: one component shortage cascades up a multi-level BOM and blocks a finished good — putting at risk a customer order worth millions, and the program milestones and contractual commitments riding on it.

Today Pelico is strong at diagnosing that risk — which order is exposed, which part is missing, why. The next frontier is prescriptive and orchestrated: recommend the best action, and align many decision-makers — human planners and, increasingly, AI agents (ours and third parties') — toward a single objective. That is an operations-research problem at its core, and it's our hardest, most defensible value.

We're hiring a Head of Optimization to found and lead the Solver squad as a player-coach. You'll start with a small founding group (3–4 engineers, including two of our deepest planning-algorithm engineers) and grow it as the impact you ship justifies it. At this stage the leadership job and the engineering job are the same job.

Why OR is central here

Pelico's projection engine is a deterministic digital twin — rules-based, reproducible, explainable (not a learned model). That property is the point: it gives optimization two things a generic agent stack cannot.

• An environment to act in: the twin is a realistic model of the plant, so a proposed action can be simulated and its downstream ripple seen before anyone commits.

• A scoreboard to act toward: the engine prices any action's global impact on a configurable North Star — OTD, turnaround time, inventory €, revenue — at the scope that matters (program / SLK, site, customer), not a local proxy.

Optimization built on that substrate is what turns Pelico from a visibility tool into a system that recommends and, under guardrails, acts.

What you'll do

Optimize allocation and recommendations — beyond FIFO. The engine allocates supply to demand by FIFO today. You'll formulate allocation as optimization under real constraints (capacity, lot sizing, safety policy, priorities), with an explicit objective rather than arrival order — and turn plan-change recommendations (pull-in, push-out, cancel, re-allocate, substitute) into optimized, actionable, partial-where-useful proposals a planner can trust.

Align actions to a single North Star. Make the North Star configurable (plus secondary KPIs), and score every action — whoever takes it — by its global impact so that local optimization converges to the company objective. Handle the hard parts: cost-aware scoring so an agent can't improve one metric by wrecking an unpriced one; hard constraints for non-negotiables (customer SLAs); lexicographic priorities vs. a monetary common denominator; and arbitration when multiple agents act on the same plan. The projection engine is the final judge — solvers propose, the twin validates and scores.

Model uncertainty. Move key inputs from point estimates to distributions — lead times, production cycle times, yield/scrap — and bring stochastic/robust methods and confidence signaling to the plan, so recommendations are honest about what the twin doesn't know.

Compute our own lead and cycle times. Declared ERP parameters are often wrong. You'll drive the ingestion of historical actuals and build the pipelines that turn them into statistical lead/cycle times and variability estimates — the foundation for uncertainty modeling, buffer recommendations, and provable accuracy.

Optimize stock and safety policy. Compute safety stock, safety time, reorder points, and lot sizes from real demand and lead-time variability at a target service level — replacing manual, static, over-conservative parameters with data-driven, auditable ones.

Make optimization agent-actionable. Package the twin and scoreboard as governed, typed, explainable, evaluated tools that agents can call with no supply-chain knowledge — read the projected plan, try an action without committing, run a larger multi-level what-if, score it against the North Star, explain the impact via causal lineage, and commit under human-in-the-loop gates. Partner with the agent-platform squads on the tool contracts and eval that let an agent safely close the loop from detection to action.

Own the trade-off between speed and optimality. There's rarely one "best" algorithm. A near-perfect plan that takes hours is useless inside an interactive tool; a fast, rough answer is sometimes good enough and sometimes not. For each problem you decide — deliberately, and able to defend it — how much optimality is worth how much runtime and memory, and you make that call hold on our largest customers (datasets of hundreds of millions of work orders, components, and events). At this scale, whether an algorithm is fast enough depends as much on how it's implemented as on which algorithm you chose, so you'll work on both alongside the engineers.

Make the results explainable and trusted. An optimizer only creates value when a planner actually acts on it. Every recommendation has to carry a clear reason — why this action, what it improves — and be checked against how the plant really behaves, so users adopt it instead of overriding it. Winning that trust is part of the job, not an afterthought.

Lead by example, and grow the team on impact. Run the squad's load-bearing OR ↔ backend-engineering pairing: author the models and specs the engineers implement and scale, so correctness lives upstream of code and nothing collapses into per-customer forks. Write models, write code, review PRs at depth, and mentor the founding engineers (deep Kotlin planning-algorithm experts) into optimization technique. Earn headcount by shipping value, then hire deliberately.

What you bring

Required

• Advanced degree in Operations Research, Applied Mathematics, or Industrial Engineering — or a track record that unambiguously demonstrates equivalent depth.

• Production experience across combinatorial optimization (MILP/LP/CP), metaheuristics, and simulation — shipped on commercial or open solvers (Gurobi, CPLEX, OR-Tools, Hexaly, or equivalent) on real problems.

• Depth in discrete-manufacturing or supply-chain planning: MRP, multi-level BOM, pegging, coverage, scheduling, allocation, or inventory under uncertainty.

• A habit of making the exact-vs-heuristic and precision-vs-speed-vs-memory trade-off deliberately, and evaluating algorithm variants rigorously.

• Large-scale performance instinct: optimization or simulation over millions-to-hundreds-of-millions of records, memory- and latency-aware.

• Hands-on in a production codebase, shipping and reviewing regularly (the squad builds in Kotlin/Java — fluency or a credible fast ramp).

• Leadership by example — you've led by being one of the strongest engineers in the room, not by leaving the keyboard.

Strongly preferred

• Planning and optimization experience in a supply-chain or manufacturing context.

• Optimization or simulation packaged for programmatic/agent consumption — typed APIs, eval harnesses, anytime profiles.

• AI-native delivery: agentic workflows, eval on model outputs, spec-driven development.

• French (an HQ benefit, not a requirement).

Why this role matters

Aligning many actors — planners and agents — on one trustworthy objective, with optimized recommendations the twin can simulate, score, and explain, is what turns Pelico from visibility into a semi-autonomous production-control copilot. As the player-coach who both invents the optimization and leads the team that ships it at scale, you own the hardest, most durable part of that bet.