HEC Professor's Deep-tech Startup Secures Top French Grant
While Bpifrance chose “culot” (“daring,” as in HEC’s motto) as the theme of its annual meeting, Praedon, a startup co-founded by HEC professor Maximilian Schiffer, has just received the highly selective Bpifrance i-Lab innovation grant. Professor Schiffer joined HEC’s faculty this September. Praedon is a successful deep-tech startup that uses AI to optimize high-stakes operational decisions.
Professors Axel Parmentier (École des Ponts ParisTech) and Maximilian Schiffer (HEC Paris), co-founders of the startup Praedon
Maximilian Schiffer, who has recently joined the Information Systems and Operations Management Department as a tenured associate professor with a core affiliation in the Hi! PARIS Center, brings not only his academic experience from TUM (Technical University of Munich, one of HEC's long-standing partners), but also AI expertise grounded in very concrete operational cases. Indeed, Maximilian Schiffer is the co-founder, along with École des Ponts ParisTech Professor Axel Parmentier, of the startup Praedon, that automates complex industrial decisions. The startup officially received the prestigious Bpifrance i-Lab Grand Prix at the maximum funding volume, i.e. 600K€, on September 16.
Bridging machine learning and operational realities
While large language models (LLMs) excel at drafting memos and generating code, relying on generative AI to run physical infrastructure is an operational hazard. A chatbot can apologize when it hallucinates; an algorithm managing an automotive or chemical plant cannot. In heavy industry, a single probabilistic blind spot—such as rescheduling an assembly line past a furnace’s throughput limit—can trigger a catastrophic, multimillion-euro plant shutdown.
Rather than relying on statistical guesswork, Praedon uses novel AI systems to automate complex decisions, deploying a "two-brain" architecture governed by Decision-Focused Learning (DFL).
Here, an upstream neural network acts as the intuitive strategist, parsing volatile contextual signals like shifting supplier lead times or spot-market price spikes in real time. Downstream, a classical combinatorial optimization solver acts as an unyielding mathematical safety inspector, translating physical realities, labor laws, and equipment tolerances into non-negotiable constraints. Because the predictive models are trained directly inside the optimization algorithm, the resulting decision is mathematically guaranteed to be feasible before a single machine moves.
This same principle is transforming volatile energy trading. As renewable power from wind and solar swings by the minute, traders rely on predictive algorithms to capture fleeting arbitrage windows. Yet standard machine learning routinely generates lucrative trades that risk overloading physical transmission lines or breaching battery discharge limits. Praedon’s integrated engine ensures that every high-speed market transaction is structurally bounded by the physical realities of the grid. By bridging machine learning with the hard guardrails of operations research, Praedon eliminates the executive dilemma between dynamic commercial speed and catastrophic operational liability.
Concrete solutions for deploying emergency service vehicles and other logistical challenges
The co-founders describe their platform as a backend decision motor. “Faced with uncertain variables such as weather events, volatile markets or logistical bottlenecks, Praedon's AI processes various scenarios. It then prescribes real-time optimal, resilient decisions that take into account all constraints, thus altogether avoiding the risk of hallucinations in critical industrial environments,” they explain in their presentation for the i-Lab innovation contest.
This type of solution applies to organizations' needs in various domains, such as energy storage, operations management, information systems, supply chain, revenue management, and soon finance. For instance, a study co-authored by Professor Schiffer shows that a combinatorial optimization-augmented machine learning pipeline that allows to learn efficient policies for ambulance dispatching and redeployment in the context of a large American city's 911 call data yields a reduction in mean response time of up to 28%, which increases the chance of survival for people with a cardiac arrest significantly.
A commercially scaling enterprise
The enterprise has transitioned from a breakthrough academic project to a scaling commercial entity. Before scaling the Software-as-a-Service (SaaS) platform, the core technology was stress-tested and proven through consulting missions and industrial pilots for global market leaders. The early validation phase, roughly from 2023 to 2025, included industrial PhD projects with Air France (where a 27% reduction in delay-induced costs was achieved), and Renault Group (where the framework generates over €1M in annual savings and cuts CO2 emissions by 10% daily across Europe).
Praedon, officially created in 2026, is now a full-blown company with active, paying customers. Through a joint venture with Flexbid, Praedon manages algorithmic intraday trading and battery storage dispatch and is scaling this smart energy trading platform across key European markets. By optimizing complex decisions, our solution generates millions of euros worth of operational savings for industry leaders. In addition, as our models use structured models, they require significantly less computing power and energy than standard LLMs.
Sources:
The research:
Schiffer, M., Hoppe, H., Su, Y., Bouvier, L., & Parmentier, A. (2026). Combinatorial Optimization Augmented Machine Learning. ArXiv.
The case studies:
Baty, L., Jungel, K., Klein, P. S., Parmentier, A., & Schiffer, M. (2023). Combinatorial Optimization enriched Machine Learning to solve the Dynamic Vehicle Routing Problem with Time Windows. ArXiv.
Jungel, K., Parmentier, A., Schiffer, M., & Vidal, T. (2023). Learning-based Online Optimization for Autonomous Mobility-on-Demand Fleet Control. ArXiv.
Rautenstrauß, M., & Schiffer, M. (2025). Optimization-Augmented Machine Learning for Vehicle Operations in Emergency Medical Services. ArXiv.
Find the Deep Tech Features on HEC's Dare Media Hub: Deep Tech: Turning Breakthrough Science into Impact