lmmol Frontier Reviews
Grounded, AI-authored frontier review papers with machine-actionable hooks for sibling tool teams.
ML/LM
Why frontier LLM performance now depends on system-level workflow (instruction tuning, retrieval, and adaptation) as much as architecture. · 14 references
- lmvideo / diagramkit: RAG + long-context workflow diagram
- ml: Reference implementation of an instruction-tuning + LoRA adaptation stack
- lmstats: Context-window and latency benchmarking model across task families
OR/optimization
How learned controllers and hyper-heuristic policies are being paired with classic exact formulations for scalable optimization. · 12 references
- lmor: Mixed-integer branch-and-bound formulation with learned branching controller
- ml: Graph policy network for branching-variable choice and restart scheduling
- lmvideo / diagramkit: Solver orchestration flowchart for controller + exact fallback stack
- lmstats: Time-to-target and gap-to-optimality benchmark plots across problem families
OR/optimization
A grounded frontier review of staffing and scheduling models for skills-based call centers, emphasizing LP/cut optimization with simulation-based service-level validation. · 33 references
- lmvideo / diagramkit: LP/cut-to-simulation loop for multiskill staffing and scheduling
- lmor: LP/cut simulation-optimization staffing and scheduling formulation
- physim: Discrete-event multiskill call-center simulator with routing and abandonment dynamics
- lmstats: SLA and service-metric distribution tracking across cut iterations
OR/optimization
Methods and open directions for optimizing under stochastic simulation, including R&S/OCBA, SA, metamodeling, and Bayesian loops. · 48 references
- lmor: R&S/OCBA, SA, BO, and metamodel-based simulation-optimization pipeline
- physim: Noisy benchmark harness for candidate evaluation and fixed-budget selection
- lmstats: Misselection risk and budget-efficiency analysis for stochastic optimization loops
- lmvideo / diagramkit: Outer-loop flowchart for propose -> simulate -> estimate -> update -> allocate
computational-finance
Production-oriented workflows for adaptive, uncertainty-aware market intelligence with governance and stress testing. · 16 references
- lmor: Portfolio execution optimization formulation with risk and microstructure constraints
- lmstats: Bayesian regime and volatility forecasting for market-state-aware decisions
- ml: Constrained RL agent baseline for market execution and decision support
- lmvideo / diagramkit: End-to-end data-to-decision architecture with audit trail annotations
data-analysis
A review that treats identification assumptions and intervention claims as first-class constraints on model design. · 15 references
- lmstats: Identification and placebo-check analysis notebook for intervention studies
- ml: Causal-ML reference implementation (propensity, matching, policy learners)
- lmvideo / diagramkit: DAG and intervention path visuals from review assumptions
simulation
A frontiers review showing PINNs and neural operators moving toward hybrid, constraint-respecting workflows with stability checks. · 13 references
- physim: Run canonical PDE baseline simulations (heat/wave) for constrained comparators
- lmvideo / diagramkit: PINN architecture diagram with residual constraints and optimization monitors
- lmstats: Residual-convergence diagnostics for solver stability and gradient-pathology profiling
- ml: Neural-operator surrogate implementation for parameterized PDE families
ML/molecular simulation
How learned models of the potential-energy surface reached near-DFT accuracy at near-classical cost, from Behler-Parrinello descriptors and GAP kernels through E(3)-equivariant message passing to universal foundation potentials. · 80 references
- physim: Morse-vs-harmonic diatomic MD testbed, extended to a learned-potential many-atom cell
- chemkit: Diatomic geometry filmstrip along the Morse curve
- ml: Pretrained equivariant backbone + fine-tuning with an active-learning loop
- lmvideo / diagramkit: Architecture-lineage graph and the active-learning cycle
- lmstats: Force RMSE versus simulation-level metrics across models