Sustainable logistics, powered by community

Good food.
Better routes.
Zero waste.

We match bakery surplus with nearby food pantries—and give volunteer drivers the smartest route to get it there fresh.

JMAKSR+42

0 neighbors
ready to drive this week

LIVE NETWORKSat, Jul 25 · Boston
BFlour & Field
PHarbor Pantry
BMorning Bell
PNorthside Shelf
BGrain House
PSt. Anne's
🚗YOU
OPTIMIZED LOOP3.2 mi · 30 min
0registered bakeries
0registered pantries
0deliveries completed this week
0all-time deliveries completed
0average pantries helped per day
VOLUNTEER TO DRIVE

One app. A meaningful trip.

Driver participation happens in the BakedBoston mobile app, so you can choose a route when it fits your day.

1✍️ Sign up

Create your driver profile.

2🪪 Verify ID

A secure check that takes about two minutes.

3📍 Choose a route

Select a nearby order now, or reserve one for a preferred route and time.

4🥖 Pick up

Drive to the bakery and collect the food.

5🥫 Drop off

Bring it to the pantry and complete the delivery.

QR code to download BakedBoston on the App StoreScan to download
the BakedBoston app
Driver routes are mobile-only. Tap or scan to open the App Store.
WHY BAKEDBOSTON

From surplus
to supper.

A simple act, coordinated intelligently.

Edible food is often discarded because a bakery cannot move it in the narrow window when it is still fresh—at the same time that local food-access organizations are under growing pressure. BakedBoston coordinates the people, places, and timing needed to turn that gap into a reliable delivery.

01
🥖

Bakeries post surplus

Pickup timing, confirmation status, and access details are added in under a minute.

02

We optimize the match

The MIP enforces complete route timing, protects expected completed pickups, then balances pantry reach and priority, raw and distributed food, equity, direct CO₂e benefit, and driver fit.

03
🚙

Neighbors drive the loop

Volunteers claim a route and deliver food while it’s still fresh.

THE OPERATIONS RESEARCH

Less waste is
a solvable problem.

At each rolling decision epoch, BakedBoston uses a Gurobi mixed-integer model to choose jointly among complete, feasible driver–bakery–pantry plans. Each route column includes a just-in-time departure, five minutes to load at the bakery, the trip to the pantry, and five minutes to unload. The formulation is a research-informed academic prototype: operational constraints are explicit, while volunteer acceptance remains a transparent synthetic assumption until real participation data exists.

Simultaneous network objectivemaximize Φ(x) = ωᵀψ(x; π),  x ∈ 𝒳

The feasible assignment x maximizes one acceptance-adjusted network feature vector ψ: pantry reach, raw and ultimately distributed food, evenness, opportunity priority, direct environmental benefit, and driver fit.

One-driver route choicer★ = arg max { πᵣ · (ωᵀφᵣ) | r ∈ ℛ(d) }

Among driver d's feasible routes ℛ(d), choose the route with the largest completion likelihood πᵣ times completed-impact value ωᵀφᵣ. There is no separate participation stage or 99%-of-best cutoff.

Why combine them?

A route that is unlikely to happen earns less expected-impact credit, but a small difference in a synthetic probability cannot automatically veto a route that rescues substantially more food or produces a fairer, cleaner result.

Completion likelihoodπᵣ = σ(2.2 − 0.045mᵣ − 1.8wᵣ − 1.1sᵣ)

mᵣ is drive time; wᵣ and sᵣ are normalized requested-time and requested-area misses. This is an auditable scenario assumption—not a trained behavioral model.

Food and direct environmental accountinghᵣ=qᵣηᵦδₚ;  γᵣ=qᵣeᵦ−[qᵣ(1−ηᵦ)eᵦ+qᵣηᵦ(1−δₚ)eₚ]−Tᵣ

Bakery-unusable food follows that bakery's landfill/pig-farm/compost mix. Pantry-undistributed food follows that pantry's unique landfill/pig-farm mix; pantry compost is zero.

ω = (10, 10, 10, 10, 10, 10, 20, 20)ᵀ. Every component in φᵣ is normalized to [0,1] before the policy weight is applied, and route contributions are adjusted by πᵣ. Hard feasibility is enforced first, so an attractive score can never override a closed window or missed deadline.

xᵣ · assignment

A binary variable: 1 when complete timed route r is selected, otherwise 0.

πᵣ · completion likelihood

A synthetic probability based on driving burden and driver time/location fit. It scales impact instead of serving as a hard preliminary objective.

κᵣ · pantry coverage · 10

Modeled completion credit for available unique pantry identities, capped once per pantry.

νᵣʳᵃʷ + εᵣʳᵃʷ · raw food · 10 + 10

Acceptance-weighted raw bakery-food volume plus evenness calculated from cumulative expected raw donations.

νᵣˢᵃᵛᵉᵈ + εᵣˢᵃᵛᵉᵈ · saved food · 10 + 10

Acceptance-weighted ultimately distributed food, where hᵣ = qᵣ · ηᵦ · δₚ, plus cumulative expected saved-food evenness.

ρᵣ · opportunity priority · 10

A pantry available but less often served receives more acceptance-adjusted credit through its smoothed recent-opportunity history.

γᵣ · direct environmental benefit · 20

The acceptance-weighted bakery no-pickup result minus bakery-unusable waste, pantry-undistributed waste, and usable-cargo tonne-kilometre transportation emissions, min–max normalized across candidates.

χᵣ · driver fit · 20

Acceptance-weighted fit: one minus the mean normalized drive-time, requested-window, and requested-area burdens.

Planned departure

Login time is the decision epoch, not a forced departure. The model schedules the latest feasible departure, so planned lead time at the origin is not penalized.

Hard constraints

Confirmed partner windows, bakery pickup deadlines, pantry latest arrivals, full travel and service timing, at most one route per driver request, and at most one assignment per bakery pickup.

Simultaneous network solve

All requests entering the same decision epoch are optimized jointly. Each request, physical driver, and bakery pickup can appear in at most one selected assignment; an open pantry may receive multiple deliveries.

Layered route menus

Conflict-free options are allocated by rank: maximize how many drivers receive recommendation #1 before allocating recommendation #2, then optimize route utility within each layer.

What remains soft

Requested time and ZIP areas influence utility and modeled acceptance. A route may fall outside them and still be recommended when it is feasible and comparatively strong.

  • Rolling-horizon joint decision epochs
  • Complete timed route-column generation
  • Hard pickup, receiving, and service feasibility
  • One acceptance-adjusted expected-impact objective
  • Opportunity-based pantry priority and cumulative food equity
  • Direct waste-pathway and transport CO₂e accounting
  • Conflict-free, coverage-first driver route menus
  • Soft requested-time and ZIP-area matching
Explore the optimization model on GitHub ↗
DATABASE-BACKED ONE-DRIVER MODEL TRACE

Follow the current routing logic in seven auditable steps.

The walkthrough requires the current registered and validated partner network. Partner identities and coordinates come from the live database; the driver request, food realization, and decision epoch are synthetic and read-only. This explanatory trace applies the current route accounting and acceptance-adjusted expected-impact selection without creating an offer; the production service solves simultaneous requests jointly with Gurobi.

CONNECTING TO PARTNER DATABASE

Loading the current bakery and pantry network…

No fictional organizations are substituted. The current Neon database quota must be restored before this database-backed demonstration can populate.