1. Core abstraction (this is 80% of OR)

Every OR problem reduces to:

Decision variables e.g.,

𝑥 𝑖 , 𝑗 = 1 x i,j ​

=1 if rider i takes order j Objective function minimize cost / maximize efficiency e.g., total delivery time Constraints each order assigned once rider capacity ≤ 3 delivery within 10 mins

If you understand this mapping, you understand OR at a functional level.

1. The 3 layers of OR (map to difficulty) Layer 1: Linear Programming (LP) continuous variables solvable efficiently

Use case:

resource allocation

Key idea:

optimal solution lies at boundary

Layer 2: Integer Programming (IP / MIP) discrete decisions (0/1, integers) NP-hard → becomes real-world hard

This is where Zepto-type problems live.

Example:

routing assignment Layer 3: Heuristics (real-world systems) “good enough, fast” because exact solution too slow

Examples:

greedy local search genetic algorithms

1. The 5 canonical problems (know these, you’re covered)

Don’t go broad—just understand these deeply:

Assignment problem → who does what Knapsack → what fits under constraints Shortest path / graph → routing Vehicle Routing Problem (VRP) → multi-agent routing (Zepto core problem) Network flow → movement through systems

1. Tooling (just enough to recognize, not master)

You don’t need Gurobi depth, just awareness:

Google OR-Tools → practical entry point PuLP / Pyomo → “write equations in Python” Gurobi / CPLEX → industrial solvers

Key idea:

You define the problem → solver finds solution

Step 2 (4–5 hours) Implement: simple assignment problem small VRP using OR-Tools

Goal: → see how problems are encoded