Bid evaluation often looks straightforward on paper: compare numbers, pick winners, apply discounts, and move on. In practice, it's messy. Packages have missing bids, discount tiers vary by bidder, spreadsheets arrive in inconsistent formats, and "lowest total cost" can't be found reliably by manual selection without missing better global combinations.

BidSense was built to solve that problem properly.

At its core, the platform treats bid evaluation as a global optimization problem. Instead of selecting winners package-by-package, it uses a MILP model to assign every package to exactly one valid bidder while accounting for bidder-specific tier discounts that depend on how many packages they win. That means the result isn't just "a good answer", it's the guaranteed lowest-cost award allocation under the rules provided.

A major part of making this usable in a real organization was making the Excel workflow strict and predictable. The backend accepts Pricing and MCA sheets in both wide and long formats, validates bidder alignment across files, rejects malformed values, and allows empties only when they're truly empty (blank/NaN/NULL). That strictness is intentional: if the optimizer is going to produce decisions worth millions, the input layer has to be uncompromising.

Once solved, BidSense generates decision-ready Excel outputs: a bidder summary, a package assignment sheet that preserves the original package order, and a detailed breakdown showing base prices, discount rates, deductions, and final prices — all formatted cleanly for stakeholders. The goal wasn't just to compute an outcome, but to produce an output that can be reviewed, shared, and trusted.

The impact was significant: this approach reduced evaluation time from hours to milliseconds and enabled the identification of true lowest-cost strategies at enterprise scale, contributing to substantial cost savings for Mubadala Energy.

On the engineering side, the most interesting part of BidSense was balancing rigor with usability. The optimizer is only as valuable as the constraints it respects and the data it can reliably consume. That's why the system focuses heavily on strict parsing, clear error messaging, and structured outputs, the kind of details that don't show up in a demo, but determine whether a tool survives real-world use.

This project also reinforced how much value lives in messy engineering: validation, reproducibility, deterministic outcomes, and clear reporting. When the result affects real procurement decisions, those qualities matter more than flashy features, and BidSense was built with that standard from the start.