Inspiration
Ithaca has a food insecurity problem that most people don't see. Friendship Donations Network rescues over 1,400 lbs of fresh food every single day — from grocery stores, bakeries, farms, and university dining halls — and redistributes it to neighbors in need across Tompkins County. They serve 2,000+ people every week with 200 volunteers and just one part-time staff member.
As ORIE students, we saw immediately that this was a logistics problem hiding inside a social impact problem. The coordinator spends hours every morning manually matching donors to pantries to drivers — by phone, by spreadsheet, by group text. Food goes uncollected. Volunteers get burned out. Pantries receive the wrong amounts at the wrong times.
We wanted to build something that directly applies our training in supply chain optimization and AI — not as a theoretical exercise, but as a real tool for a real organization doing real work in our own backyard. The intersection of operations research, artificial intelligence, and community impact felt like exactly the kind of problem we came to Cornell to solve.
What it does
FDN Route Optimizer is an AI-powered logistics coordinator that solves the daily matching problem for Friendship Donations Network in seconds rather than hours.
A coordinator opens the app and either types a plain-English description of the day's donations or uploads a structured spreadsheet. Claude parses that input into structured data — donor names, quantities, locations, and expiry windows. The optimizer then assigns donations to volunteer drivers and pantries using a provably sound operations research algorithm, generating a route plan that respects every constraint simultaneously.
The output includes an interactive map showing each driver's road-following route, a Claude-generated plain-English briefing each driver can actually read, a resource dashboard showing driver capacity utilization and pantry status in real time, and a before/after comparison that quantifies exactly how much better the optimized plan is versus naive sequential assignment.
If a donation can't be routed, the dashboard explains exactly why — time window missed, driver at capacity, pantry closed — so the coordinator can make an informed decision rather than wonder what went wrong.
How we built it
We started with a rigorous problem formulation before writing a single line of code. The core of the optimizer is a Vehicle Routing Problem with Time Windows (VRPTW), one of the most well-studied problems in combinatorial optimization.
Formal formulation:
Let x_ijk = 1 if driver k travels from stop i to stop j, and t_ik denote the arrival time of driver k at stop i.
Objective:
min sum_k sum_i sum_j d_ij * x_ijk + lambda * sum_p u_p * (1 - y_p)
Subject to:
Time windows:
a_i <= t_ik <= b_i for all pantry stops i
Driver availability:
avail_start_k <= t_ik <= avail_end_k for all drivers k
Vehicle capacity:
sum_i load_i * x_ijk <= cap_k for all k
Perishability constraint:
t_ik <= T_p
Since VRPTW is NP-hard, we implemented a greedy insertion heuristic with Earliest Deadline First (EDF) priority ordering — the right engineering choice for our problem scale, where near-optimal solutions are found in milliseconds.
The tech stack came together around that core. We used Claude API for the natural language interface — two carefully engineered prompts handle input parsing and output narration. OpenRouteService provides real pairwise travel time matrices between Ithaca locations. Folium renders road-following routes on an interactive map. Streamlit gave us a clean multi-page app — Home, Resources, and Optimizer — without needing a separate frontend framework. All data is grounded in real FDN pantry addresses, time windows, and partner organizations sourced directly from friendshipdonations.org.
Challenges we ran into
The first hour of the hackathon was spent entirely on setup — AJ had a new computer and needed to reinstall Python, conda, Git, and all dependencies from scratch before writing a single line of code. Not the ideal start to a 6-hour sprint.
Midway through, AJ's machine lost access to GitHub through VS Code entirely. Rather than spend time debugging OAuth tokens, we adapted our division of labor on the fly — AJ pushed changes through the terminal while Tanushree managed the VS Code integration. It slowed us down but kept both partners productive.
On the product side, the hardest challenge was narrowing from a broad idea — "food redistribution optimization" — to a concrete, buildable scope. There were many directions the project could have gone: demand forecasting, donor matching, volunteer scheduling. Committing to the VRPTW formulation early was the decision that unlocked everything else.
Shoutout to Claude for being an incredibly patient third teammate throughout — from ideation and formulation all the way through debugging and production.
Accomplishments we're proud of
This was the first official hackathon for both of us — and we shipped something real.
We're proud that the project directly applies our ORIE training rather than using it as decoration. The EDF heuristic, the VRPTW formulation, the feasibility checker — these aren't buzzwords in our submission, they're running code.
We're proud that we built on real data. Real pantry addresses. Real time windows. Real Ithaca streets. This isn't a demo with fake coordinates — it's a tool that could run tomorrow morning.
Most of all, we're proud that we built something that could genuinely help. Tompkins County food insecurity is real, FDN's coordination burden is real, and the time we could give back to a single overworked coordinator is real.
What we learned
Claude is a genuine force multiplier for end-to-end project completion. From helping us frame the problem, to writing and debugging code, to generating the driver briefings inside the app itself — it compressed what would have been weeks of work into a single day.
Food insecurity in Ithaca is more prevalent and more operationally complex than we expected. FDN serves over 50 community programs across Tompkins County with a rotating weekly schedule that changes every day. The logistics challenge is real and the stakes are high.
We deepened our familiarity with GitHub collaboration under pressure, and picked up practical experience with libraries we hadn't used before — Folium, Streamlit, OpenRouteService, and the Anthropic SDK.
And as Tony Chen said — Claude makes it so that the only limit to what we can create and improve in the world is our imagination. Today felt like proof of that.
What's next for FDN Route Optimizer
In building this tool, we reached out to Friendship Donations Network directly to tell them about our work and express our admiration for what they do. We hope that by sharing the tool with them, we can earn real feedback from the people who would actually use it every day.
The next version would incorporate coordinator feedback, real-time donor notifications, and a more sophisticated matching engine that accounts for pantry-specific food preferences and recipient demographics. We'd also want to add volunteer onboarding — a way for new drivers to register their availability directly through the app rather than through a coordinator.
Beyond FDN, the underlying engine is fully generalizable. Any food rescue network operating on a similar neighbor-to-neighbor model — in Burlington, Binghamton, or any small city — could deploy a version of this tool with minimal adaptation. That's the version of this project we want to build next.
Log in or sign up for Devpost to join the conversation.