The Full Story
Technical architecture, research methodology, government use cases, measurement framework, and roadmap.
detect, inform, respond the full loop
FreshFind is a three-layer AI system that gives governments something they've never had: predictive food security intelligence. It watches the signals, identifies who's at risk, and deploys the response. Continuously. Autonomously.
Detect: FreshFind's Barometer identifies a 35% spike in financial stress indicators across eastern Kentucky zip codes 41271-41274. Simultaneously, the policy parser flags that expanded SNAP work requirements take effect in 60 days.
Inform: The system generates a briefing for the Kentucky SNAP director: 2,400 households estimated at risk. 60% of current recipients ages 55-64 in these zip codes. Nearest fresh produce: 12 miles. Recommended: activate targeted outreach, coordinate mobile pantry deployment, alert 211 center for surge capacity.
Respond: The state agency sends personalized notifications through their existing channels. Recipients land in FreshFind's guided intake. Cross-program eligibility check. Food access routing. Application assistance. Families connected to help before the policy change takes effect.
This is not a benefits navigator
Most AI-for-government projects in the benefits space focus on the administrative layer: helping caseworkers process applications faster. That's a different problem. FreshFind gives governments a capability they've never had: predictive intelligence that sees food crises forming and orchestrates a response before they peak. The user isn't a person searching for food help. The user is a state SNAP director, a county emergency manager, a health department administrator deciding where to deploy resources.
technical architecture under the hood
A structured pipeline of specialized models. Each stage produces auditable, deterministic outputs. No black boxes. A state SNAP director can trace any recommendation to the specific data source and rule that generated it.
the response layer is already built live now
The Detect and Inform layers are what the grant builds. But the Respond layer is already functional. The grant connects these tools to the intelligence system that tells governments where and when to deploy them.
Financial Health Barometer
Real-time financial stress tracking across all 50 states. Data from BLS, Census, HUD, FRED, and Google Trends. This is the signal layer that powers the entire Detect stage. It already exists. The grant adds the AI that turns signals into predictions.
Food Desert Analyzer
Enter a zip code. See whether your neighborhood has access to grocery stores, fresh food, and transportation. Powers the food access component of every intelligence briefing.
Food Assistance Calculator
Cross-program eligibility screening. One intake covers SNAP, WIC, school meals, food pantries, utility assistance, and more. Deployed by governments into communities where the intelligence layer identified the need.
calibrated on ground truth not assumptions
Predictive models are only as good as their calibration data. We spent two years in five communities mapping what's actually happening: who's food insecure, why programs aren't reaching them, what's blocking access. This is the ground truth that makes FreshFind's predictions credible. It's also why these five communities are our pilot targets.
built for government users
The primary user is a government decision-maker. A state SNAP director. A county emergency manager. A health department administrator. FreshFind gives them predictive intelligence and response tools. They deploy through channels people already trust.
Our pilot strategy starts in the five communities where we have two years of field relationships and research data. We are in active conversations with county health departments in Wolfe County, KY and Twin Falls, ID, and with food bank networks in Chicago and Atlanta, to formalize pilot partnerships. These aren't cold introductions. Our research with the University of Virginia was conducted in collaboration with local agencies in these communities.
State SNAP & Medicaid Agencies
State agencies are legally required to notify affected households when policy changes hit. Most have no automated system to do this at scale. FreshFind is that system. The Detect layer flags the change. The Inform layer identifies affected households. The agency sends personalized outreach under their name. Respondents are routed into FreshFind's response tools.
The agency gets a dashboard: households reached, alternative program enrollments, food resources accessed, food access score changes in target zip codes. Real outcomes, not just process metrics.
County Emergency Managers
When a disaster declaration drops, FreshFind automatically identifies affected populations and activates D-SNAP screening, emergency food routing, and crisis-specific program recommendations. The emergency manager gets a briefing within hours: who's affected, what's available, where the food access gaps are.
211 Centers, Health Departments, School Districts, Libraries
Response deployment channels. 211 operators use FreshFind as real-time support. Health departments run 2-minute screenings during WIC visits. School districts use free lunch data to identify families who qualify for more. Libraries host kiosk-style terminals. The intelligence layer tells governments where to focus. These channels deliver the help.
What Government Partners Get
Data & Security: No user data sold or retained. Google Cloud with FedRAMP-authorized, HIPAA-compliant infrastructure. Encrypted at rest and in transit. Government partners control their data. SOC 2 Type II audit Month 4.
how we'll measure it
built to replicate
The Barometer already covers all 50 states. Adding a government partner means pointing the intelligence layer at their geography and plugging in local food access data. The scaling unit is the government partnership. One state's success becomes another state's blueprint.
Every presidential disaster declaration is an automatic scaling trigger. 60+ per year in the U.S. Each activates the full Detect-Inform-Respond loop without manual intervention.
The model travels. FinMango operates in 13+ countries. India's Public Distribution System, Brazil's Bolsa Família, EU food assistance programs. Every country with social benefit programs has the same blind spot. At the end of the 18-month grant, we publish a full open-source toolkit for national and international replication.
the roadmap
$1.5M over 18 months. The Barometer and response tools are live. The grant builds the intelligence layer that connects them.
Scott Glasgow · Sarah Cherian · Soham Patel · Tony Ramos
FinMango · finmango.org
