← Thinking Out Loud
Decision Optimisation Radar

Daily signals from the frontier of decision optimisation.

Daily signals in decision optimisation: industry moves, research papers, and tools scanned every morning across arXiv (Cornell's open-access preprint archive), research labs, solver vendors, and domain journals. Alongside a standing Tool Library, a growing index of DO Concepts, and an events calendar for conferences, workshops, and competitions.

Dailyautomated scan 30+sources OR · MIP · CP · LLM+OR · +more
Supply Chain · Healthcare · Energy · Finance · Transport · Manufacturing · +more
DO Concepts Tool Library Events & Conferences
RSS
Archive — newest first
12 May 2026 · 2 industry 2 papers 1 case study 1 term

Decision Optimisation Radar — 12 May 2026

Industry
  • E-CVRP LAHC. Instance-aware parameter prediction cuts routing cost 0.28—no extra run-time overhead
  • GEM gap-filling (CSIRO). Metaheuristic across all growth conditions beats ILP baseline on every metric
  • PuLP 3.3.1. Drop-in patch; compatible with HiGHS 1.14 and Gurobi 13
Research
  • Linear Decision Trees. Offline-synthesised policy returns ILP optimal solutions orders of magnitude faster on repeated queries
  • T20 Cricket MDP. Markov chain on 1,161 IPL records lifts Mumbai Indians’ win probability 4.1 pp
Case Study
  • LDT for ILP. Fixed feasible set + varying cost vector — decision tree synthesis eliminates repeated branch-and-bound
Term Bilevel Optimisation — An optimisation programme with another optimisation programme nested inside it as a constraint: the feasible set contains only solutions where the follower is also optimising.
Metaheuristics Decision Trees ILP Bilevel Sports Analytics
8 May 2026 · 1 industry 1 papers 1 case study 1 term

Decision Optimisation Radar — 8 May 2026

Industry
  • NVIDIA cuOpt Agent Skills. GPU-accelerated vehicle routing exposed as LangChain-native tools so AI agents can call a solver the same way they call any other function.
Research
  • DRCIRP (arXiv:2605.03785). Cyclic inventory routing under distributional uncertainty reformulated as a tractable MISOCP via moment-based duality.
Term Ambiguity Set — A bounded collection of probability distributions consistent with observed data moments; the plan must be optimal against the worst distribution in the set.
GPU Solvers Inventory Routing Distributionally Robust Ambiguity Set Agent Orchestration
5 May 2026 · 2 industry 2 papers 1 case study 1 term

Decision Optimisation Radar — 5 May 2026

Industry
  • UPMC anesthesiologist staffing. Three-stage MILP across 11 hospitals and a shared on-call pool nets $8,382/day ($800K+ annually) against ad-hoc scheduling.
  • ECCO Shoes stochastic replenishment. Two-stage stochastic MIP with divide-and-conquer decomposition automates 300,000 monthly orders across 536 stores, cutting cost 1.09%.
Research
  • PyVRP+ time windows & mixed fleet. Open-source HGS solver extended with time-window feasibility enforcement and heterogeneous-fleet support, matching best-known VRPTW benchmarks.
  • Sparse branching for MIP. Bayramoglu, Nemhauser & Sahinidis show sparse problem-structure features match strong-branching accuracy across MIPLIB families at near-zero per-node cost.
Case Study
  • ECCO two-stage stochastic decomposition. Divide-and-conquer splits 536-store global inventory allocation into country subproblems coordinated by shared global inventory constraints.
Term Dual Decomposition — Duplicates shared variables across independently-solved subproblems and enforces agreement through iteratively-updated Lagrange multiplier prices — the stochastic-programming instantiation is known as progressive hedging.
Healthcare Decomposition Stochastic Vehicle Routing MIP
3 May 2026 · 2 industry 2 papers 1 case study 1 term

Decision Optimisation Radar — 3 May 2026

Industry
  • Air France stochastic tail assignment. Column-generation MIP with stochastic pricing achieves 0.28% optimality gap on 600-leg instances, outperforming deterministic minimum-turn-time baselines.
  • Anna Chakra — India PDS. Network-flow OR deployed across 30 states saves ₹2.5 billion annually, cuts logistics distances 15–50%, and reduces emissions 35% for 810 million beneficiaries.
Research
  • RKHS uncertainty sets for DRO. Kernel-based uncertainty sets harden stochastic programmes against out-of-distribution shifts where Wasserstein and moment sets fail.
  • RL-guided Benders decomposition. RL agent + KKT-informed neural cuts reduce MINLP solve time 57.5% versus classical Generalised Benders Decomposition.
Case Study
  • Anna Chakra — multi-commodity network flow. Hierarchical decomposition: central-to-state LP allocation then capacitated vehicle routing per district, repeated monthly across all 30 states.
Term Time-Space Network — A directed graph where nodes are (location, time) pairs and arcs encode movement or waiting — transforming scheduling and routing problems into classical network-flow problems solvable by minimum-cost flow and shortest-path algorithms.
Network Flow Stochastic Decomposition Public Sector OR Tail Assignment
2 May 2026 · 1 industry 2 papers 1 case study 1 term

Decision Optimisation Radar — 2 May 2026

Industry
  • Jacobi Robotics + Delta Tech. OmniPalletizer deployed live at a defence manufacturer: 90% cube utilisation, 100% pallet stability, two-week commissioning on fully random mixed-case flow — zero gap between simulation and production floor.
Research
  • ML-guided MBQP heuristics. Machine learning classifier predicts variable fixings for primal heuristics inside branch-and-bound, outperforming Gurobi defaults on mixed-binary quadratic program benchmarks.
  • CP for FJSP + robots. Constraint programming model encodes flexible job shop scheduling and mobile robot routing jointly, eliminating the makespan penalty of sequential decomposition.
Case Study
  • ML primal heuristics on MBQP. Classifier trained on solved instances guides variable fixing inside branch-and-bound to find incumbents faster than hand-designed heuristic rules.
Term Arc Consistency — Every variable value must have a compatible partner in every directly constrained neighbour; classical arc consistency covers binary constraints, Generalized Arc Consistency (GAC) extends this to global constraints like AllDifferent and NoOverlap.
Mixed-Binary QP ML + MIP Constraint Programming Job Shop Scheduling Robotic Palletizing Arc Consistency
30 Apr 2026 · 1 industry 2 papers 1 case study 1 term

Decision Optimisation Radar — 30 April 2026

Industry
  • NVIDIA cuOpt. First GPU solver to win LP Feasibility category on the Hans Mittelmann leaderboard — GPU interior-point methods now competitive with CPU incumbents.
  • SDDP.jl v1.13.2 + SCIP 10.0.2. PAR(k) tutorial lands in SDDP.jl; SCIP 10.0.2 ships cut-based conflict analysis improvements.
Research
  • Benders Filtering. Cut filtering by violation magnitude reduces active Benders pool by an order of magnitude on decomposable mixed-integer linear programs.
  • Surgical MILP. Explicit overtime cost in MILP formulation produces provably near-optimal operating room schedules on real regulated hospital instances.
Case Study
  • cuOpt LPFeas. GPU infrastructure deployed for machine-learning training can serve LP feasibility workloads without solver change.
Term Conflict Analysis — A MIP solver converts infeasibility proofs from branch-and-bound sub-problems into reusable constraints, permanently pruning dead regions of the search tree.
LP Feasibility GPU Solvers Benders Decomposition Surgical Scheduling Conflict Analysis
29 Apr 2026 · 2 industry 2 papers 1 case study

Decision Optimisation Radar — 29 April 2026

Industry
  • AFRY BID3 — FICO ESG Champion 2026. Branch-and-cut unit commitment on the Great Britain grid: cutting planes on minimum up/down times make 48-hour renewable-integrated dispatch tractable in seconds.
  • Erste Group — 22% profit lift in retail lending. MILP replaces 90% manual exception pricing at one of Central and Eastern Europe's largest banks; knapsack cover cuts enforce regulatory capital constraints at portfolio scale.
Research
  • GPU-accelerated branch-and-cut for unit commitment. Parallel cut generation on NVIDIA A100 reduces solve time for 48-hour UC horizons by 8× over CPU Xpress baseline.
  • Gomory–Chvátal cuts for loan portfolio optimisation. Structured cut families derived from regulatory balance-sheet constraints close 80%+ of the integrality gap at the root node, eliminating most branching.
Case Study
  • Erste Group retail lending MILP. Full formulation: MILP over rate × segment decisions, MIR cuts on capital constraints, 22% profit lift, near-zero manual exceptions in production.
Term Cutting Planes — valid linear inequalities added to the LP relaxation that slice off fractional corners without removing any integer-feasible solution, steadily tightening the bound toward the integer hull.
Energy Financial Services Branch-and-Cut Unit Commitment Lending Optimisation
27 Apr 2026 · 2 industry 2 papers 1 case study 1 term

Decision Optimisation Radar — 27 April 2026

Industry
  • Kinaxis + NVIDIA cuOpt. GPU-parallel LP/MILP/VRP solver cuts semiconductor supply chain planning cycle from 3+ hours to 17 minutes across 250K scenarios/month for 400+ enterprise customers.
  • Timefold Solver 2.0. New Neighborhoods API enables custom Large Neighbourhood Search moves without forking the solver; list variables add native routing and scheduling sequence support.
Research
  • NBA roster MIP under salary-cap uncertainty. Rolling-horizon distributionally robust MIP optimises five-year franchise roster decisions under multi-year salary-cap ambiguity, validated on 500 simulated seasons.
  • T20 cricket MDP: joint batting + bowling. Stochastic dynamic programme co-optimises batting order and bowling plan, improving win probability by 4.1 percentage points on two 2026 Indian Premier League matches.
Case Study
  • Kinaxis Maestro: 12× speedup deep-dive. GPU-parallel solver substitution eliminates overnight batch constraints without changing the underlying LP/MILP planning model, enabling real-time scenario evaluation.
Term Bellman Equation — the recursion that sets a state's value equal to the best immediate reward plus the discounted value of the next state — the self-consistency condition that is all of dynamic programming.
GPU Acceleration Supply Chain Dynamic Programming Sports Analytics MDP
26 Apr 2026 · 2 papers 1 case study 1 term

Decision Optimisation Radar — 26 April 2026

Research
  • VRP temporal dependency. Fragment-based price-cut-and-enumerate algorithm beats all benchmarks on vehicle routing with ordered visit constraints across healthcare, aviation, and field service.
  • Robust scheduling: time-of-use costs. Fixing job order before uncertain electricity tariffs are revealed outperforms joint optimisation; budgeted uncertainty set keeps the robust counterpart tractable.
Case Study
  • VRP-TDC formulation deep-dive. Fragment-based reformulation compresses the variable count and enables exact solutions on real-scale instances that defeat arc-flow approaches.
Term Budgeted Uncertainty Set — a constraint capping how many uncertain parameters can simultaneously take worst-case values, controlled by a budget Γ that sits between the deterministic point and the full worst-case box.
Routing Robust Optimisation Scheduling Uncertainty Modelling Solver Infrastructure
24 Apr 2026 · 1 industry 1 papers 1 case study 1 term

Decision Optimisation Radar — 24 April 2026

Industry
  • Anna Chakra — Edelman 2026. India’s DFPD and IIT Delhi deploy OR-based route optimisation across the national PDS supply chain, saving ₹250 crore per year for 810 million beneficiaries.
Research
  • Pseudo-compact MIP for stochastic CVRP. First family of closed-form MIP formulations for CVRPSD solvable in Gurobi or HiGHS without scenario enumeration — Tomazella, Munari, Morabito.
Case Study
  • Anna Chakra formulation. Minimum-cost network routing over India’s multi-echelon foodgrain network: objectives, constraints, and what made national-scale data integration the hard part.
Term Predict-then-Optimize — A two-phase ML-to-solver pipeline whose modular elegance hides a training-objective mismatch: MSE-trained forecasts can produce suboptimal decisions when prediction errors fall in solver-sensitive directions.
Route Optimisation Stochastic VRP Public Sector OR Predict-then-Optimize Edelman 2026
23 Apr 2026 · 1 industry 2 papers 1 term

Decision Optimisation Radar — 23 April 2026

Industry
  • Eurowings crew pairing optimizer. First airline to deploy Lufthansa Systems’ renewed cloud-native crew pairing application, moving column-generation-based crew scheduling into elastic cloud compute.
Research
  • LP algorithm benchmarks. Simplex vs IPM vs PDHG: asymptotic scaling diverges sharply by application class across six LP problem families.
  • PyVRP+. LLM metacognitive layer auto-evolves heuristic operators inside Hybrid Genetic Search, closing gaps on CVRP, VRPTW, and EVRP without manual tuning.
Term Second-Order Cone Programming — Optimises a linear objective over cone constraints, subsuming LP and convex QP, solvable to global optimality in polynomial time — the natural framework for portfolio CVaR bounds, power flow relaxations, and Wasserstein-ball distributionally robust optimisation.
Crew Planning Linear Programming Metaheuristics Conic Optimisation
22 Apr 2026 · 2 industry 2 papers 1 case study 1 term

Decision Optimisation Radar — 22 April 2026

Industry
  • ECCO Shoes + INFORMS Edelman. Two-stage stochastic MIP co-designs seasonal assortments and store allocations over millions of SKU-store-week cells using scenario generation plus column generation over assortment templates.
  • Google carbon-aware compute. Multi-objective MIP fuses real-time grid carbon forecasts with datacentre capacity, shifting batch workloads across regions and hours for multi-million-tonne CO₂ reduction at under 1% latency cost.
Research
  • Exact VRP with Temporal Dependencies. Arc-flow branch-and-cut with synchronisation, precedence, and bounded-lag cuts closes over 90% of instances within an hour at fleet sizes up to 50 vehicles.
  • Rolling-Horizon DRO-CVaR for NBA Rosters. Multi-season stochastic MIP with a Wasserstein ambiguity set delivers 4–6% cap-efficiency gain and 40% tail-cost reduction over risk-neutral baselines on twelve seasons of public contract data.
Case study
  • ECCO Interactive Assortment Recommender. Two-stage stochastic MIP pipeline — scenario generation feeds second-stage allocation cost back into a column-generation master that prices new assortment templates against first-stage decisions.
Term Distributionally Robust Optimisation — worst-case expected cost when the disturbance distribution itself is uncertain and constrained only to lie within an ambiguity set — between classical Robust Optimisation's fixed set and Stochastic Programming's single known distribution.
Supply Chain Energy Transport Sports Stochastic
21 Apr 2026 · 2 industry 2 papers 1 case study 1 term

Decision Optimisation Radar — 21 April 2026

Industry
  • TraceLink + Kinaxis. Multienterprise trading-partner execution feeds pipe live purchase-order and ASN state directly into Maestro concurrent planning.
  • Mass General + GE HealthCare. AI missed-opportunity-risk model flags 96% of OR cases at risk of cancellation or slippage, driving sub-hourly theatre reallocation.
Research
  • Inexact Trust-Region CVaR. Approximate subgradient oracle inside a trust-region loop cuts solve time 2-5x on risk-averse stochastic programmes with CVaR objectives.
  • Active Constraint Acquisition. CP scheduler that learns its missing rules from operator accept-or-reject feedback closes a 14% revisit-rate gap in under 20 queries.
Case study
  • Earth-observation satellite scheduling. Propose-solve-ask-learn loop wraps a CP core with an acquisition policy and operator oracle; constraint network stabilises inside twenty queries.
Term Conditional Value at Risk — the average loss across the worst alpha fraction of outcomes — the mean of what lies beyond Value at Risk's threshold, not the threshold itself, which is what makes it coherent where Value at Risk is not.
Supply Chain Healthcare Stochastic Satellite Risk
20 Apr 2026 · 2 industry 2 papers 1 case study 1 term

Decision Optimisation Radar — 20 April 2026

Industry
  • Verusen MRO award. SDCE 2026 Pros to Know puts agentic constrained optimisation on the board for maintenance-repair-operations inventory.
  • Krones × NVIDIA digital twin. PhysicsNeMo neural CFD surrogates cut beverage-line fluid simulation cost 95% inside an Omniverse twin — with a case study teardown below.
Research
  • Mix-CALADIN consensus. Augmented-Lagrangian decomposition extended to coupled Mixed Integer Programmes with finite-time stationarity at thousand-agent scale.
  • Copositive EV pricing. Bilevel EV-charging price-setting reduces to a tractable copositive programme within 1.5% of the true optimum.
Case study
  • Krones physics surrogate stack. Five-stage pipeline from sensor + CAD input through PhysicsNeMo surrogate to optimisation-in-the-loop planner.
Term Augmented Lagrangian — a Lagrangian augmented with a quadratic stabiliser so the multiplier updates converge at finite penalty weight ρ.
Supply Chain Manufacturing Decomposition Energy Bilevel
19 Apr 2026 · 2 industry 2 papers 1 case study

Decision Optimisation Radar — 19 April 2026

Industry
  • Microsoft IFS deep-dive. Edelman-winning fulfilment stack: ML forecaster + MIP solver + OptiGuide GenAI assistant.
  • Timefold Q1 2026. Field-service release ships area preferences, same-address grouping, and labour-law breaks.
Research
  • Benders IL. Feasibility-aware imitation learning accelerates stochastic MILP Benders by 1.5–3x.
  • Satellite ACA. Active constraint acquisition converges on near-CP-SAT feasibility from operator feedback.
Term Chance Constraint — a requirement that must hold with probability 1 − α rather than in every scenario.
Supply Chain Field Service Decomposition Aerospace Stochastic
18 Apr 2026 · 3 papers 1 case study

Decision Optimisation Radar — 18 April 2026

Research
  • SRI for VRP. Scenario recourse inequalities solve 329 more two-stage routing instances.
  • Technician routing. Column generation with lexicographic optimisation on French utility real data.
  • Urban freight RL. Constraint-aware policy learning enforces feasibility during training.
Term Scenario Recourse Inequality — captures which decisions remain fixed before stochastic resolution.
Transport Routing Stochastic
17 Apr 2026 · 2 industry 2 papers

Decision Optimisation Radar — 17 April 2026

Industry
  • Edelman win. Microsoft's Intelligent Fulfillment Service takes the 2026 Franz Edelman Award.
  • NVIDIA cuOpt 26.04. Ships in GAMS with PDLP precision control and MIP heuristic tuning.
Research
  • Robust scheduling paper. NP-hardness for single-machine scheduling under budgeted uncertainty.
  • Multistage stochastic paper. Multilevel Monte Carlo breaks the exponential scenario barrier.
Term of the DayBudgeted Uncertainty Set.
Microsoft + IFS Edelman cuOpt 26.04 / GAMS Robust Scheduling MCCO / MLMC Budgeted Uncertainty Set
→
16 Apr 2026 · 1 industry 2 papers

Decision Optimisation Radar — 16 April 2026

Bridgestone deploys Hexaly Mixed Integer Linear Programming (MILP) for multi-plant tire production planning, benchmarked against Gurobi. A bilevel metaheuristic separating routing from charging sets 9 of 10 new best solutions on the IEEE WCCI-2020 electric vehicle routing benchmark. A robust portfolio approach delivers closed-form hedge ratios outperforming dynamic hedging on a nine-year multi-asset backtest. Term of the Day: Bilevel Optimisation.

Bridgestone + Hexaly Bilevel LAHC ECVRP Robust Portfolio Hedging Bilevel Optimisation
→
15 Apr 2026 · 2 industry 2 papers 9 conferences

Decision Optimisation Radar — 15 April 2026

Quickbase acquires Solvice, bringing constraint-based routing and scheduling into a no-code enterprise platform for ~12,000 customers. NVIDIA cuOpt v26.04 extends GPU LP to lower-cost FP32 hardware. Graph-RHO cuts flexible job shop scheduling time by 30%+ via critical-path-aware training. A VPP day-ahead paper achieves 100x speedup over the standard method via structural reformulation. Term of the Day: Rolling Horizon Optimisation.

Quickbase + Solvice NVIDIA cuOpt v26.04 FJSP Graph Neural Network VPP Stochastic Robust Rolling Horizon Optimisation
→
14 Apr 2026 · 2 industry 2 papers 7 conferences

Decision Optimisation Radar — 14 April 2026

Oracle named Leader in both 2026 Gartner Magic Quadrant reports for Supply Chain Planning, including the inaugural Process Industries edition. IonQ and Einride publish a hybrid quantum-classical VRP workflow improving shipments by 12% on real freight data. ReSched (ICLR 2026) outperforms all prior dispatching rules on flexible job shop scheduling. Term of the Day: Large Neighbourhood Search.

Quantum-Classical VRP Job Shop DRL EV Fairness Large Neighbourhood Search ICLR 2026
→
13 Apr 2026 · 1 industry 2 papers 3 conferences

Decision Optimisation Radar — 13 April 2026

Six 2026 Franz Edelman Award finalists present live OR deployments at INFORMS Analytics+, including ECCO Shoes' stochastic mixed-integer program for 536 stores and India's Anna Chakra food distribution system serving 810 million people. Two papers: end-to-end learning of correlated operating reserve requirements cuts energy dispatch cost by 4.8%; differentiable initialisation warm-starts ILP solvers for scheduling with 10x speedup. Term of the Day: Non-Anticipativity.

Edelman 2026 Operating Reserve Learning Differentiable Scheduling Non-Anticipativity
→
12 Apr 2026 · 2 industry 2 papers 2 conferences

Decision Optimisation Radar — 12 April 2026

HiGHS v1.14.0 extends the free HiPO interior-point solver to convex quadratic programs; Gartner forecasts Supply Chain Management software with agentic AI will reach $53 billion by 2030. Two papers address tri-level resilience optimisation for wildfire-prone power grids and a new primal heuristic for Mixed-Integer Bilevel Optimisation. Term of the Day: Big-M Formulation.

HiGHS v1.14.0 Agentic AI SCM Wildfire Grid Resilience Bilevel Optimisation Big-M Formulation
→
11 Apr 2026 · 1 industry 2 papers 4 conferences

Decision Optimisation Radar — 11 April 2026

Gartner's inaugural discrete-industries supply chain planning Magic Quadrant changes which vendors OR practitioners should benchmark; two arXiv papers show that explicit priority ordering, whether lexicographic objectives in routing or sparse features in MIP branching, outperforms implicit weighting across energy and solver-infrastructure domains. Term of the Day: Lexicographic Optimisation.

Lexicographic Optimisation Supply Chain Planning MIP Branching Column Generation Gartner MQ
→
10 Apr 2026 · 4 industry 3 papers 4 conferences

Decision Optimisation Radar — 10 April 2026

As INFORMS Analytics+ opens tomorrow, four signals track AI moving from recommender to executor: Microsoft Physical AI blueprints, constrained agents making live routing decisions at UPS and FedEx, C3.ai launching tariff-triggered scenario replanning, and Trimble confirming AI route optimisation has crossed the production threshold. Term of the Day: Integrality Gap.

Integrality Gap Physical AI Supply Chain AI Constrained Agents Tariff Resilience
→
9 Apr 2026 · 4 industry 3 papers 4 conferences

Decision Optimisation Radar — 9 April 2026

As LLMs automate MIP formulation and GPUs accelerate solving, Column Generation remains the decomposition insight that separates scalable production OR from models that stall at enterprise scale. Term of the Day: Column Generation.

Column Generation LLM+OR Supply Chain AI Healthcare Scheduling Portfolio Optimisation
→
8 Apr 2026 · 4 industry 3 papers 4 conferences

Decision Optimisation Radar — 8 April 2026

As tariff volatility and energy grid disruption mount, signals converge on a single capability: decision systems designed to stay feasible when averages fail. Term of the Day: Two-Stage Robust Optimisation.

Robust Optimisation Supply Chain AI Fleet Scheduling Energy Systems Digital Twin
→
7 Apr 2026 · 4 industry 3 papers 4 conferences

Decision Optimisation Radar — 7 April 2026

This week's dual OR conference season, INFORMS Analytics+ and AGIFORS convening simultaneously, marks the moment when decision intelligence transitions from annual planning exercise to always-on operational capability. Term of the Day: Pareto Optimality.

LLM + OR Multi-Objective Optimisation VRP GPU Optimisation Airline Operations
→
6 Apr 2026 · 4 industry 3 papers 3 conferences

Decision Optimisation Radar — 6 April 2026

As AI systems gain the ability to verify their own outputs against executable solvers and operational data, the boundary between AI-as-recommender and AI-as-executor is dissolving — verifiability is becoming the new quality floor for enterprise decision intelligence.

Verifiable AI LLM + OR Supply Chain Robust Optimisation Stochastic MPC
→
5 Apr 2026 · 4 industry 3 papers 3 conferences

Decision Optimisation Radar — 5 April 2026

The infrastructure layer for decision intelligence is maturing from research prototype to production-grade toolchain. Google DeepMind's Apache 2.0 Gemma 4 and Microsoft's RL-driven agentic AI in Dynamics 365 signal that reasoning-capable LLMs are arriving in enterprise workflows at scale without licensing friction.

LLM + OR Agentic Planning Supply Chain GPU Optimisation Combinatorial Optimisation
→
23 Mar 2026 · 5 industry 5 papers 5 conferences

Decision Optimisation Radar — 23 March 2026

The LLM-as-formulator, solver-as-executor stack is reaching production readiness. Kinaxis + cuOpt delivers 12× speedup on 50M-variable models. Gartner's inaugural DI Magic Quadrant signals category maturity. cuOpt open-sourced under Apache 2.0.

Supply Chain LLM + OR GPU Optimisation Gartner MQ Carbon-Aware Scheduling
→

Content is researched and written with AI assistance and may contain errors — always verify claims against the linked primary sources before acting on them. Audio briefs are generated using NotebookLM.