BenefitFlow

A conversational AI front-end for Minnesota SNAP. We transform the "Paperwork Nightmare" into a conversation in 28 languages, including Somali and Hmong. $30B goes unclaimed we’re bringing it home.

Comment

Inspiration

In 2026, Minnesota is facing a "Participation Gap." While the state has a 200% Federal Poverty Level eligibility threshold—one of the most generous in the nation—thousands of residents in the Twin Cities go hungry. We met "Jordan," a UMN student living on $4.22. Like many in the Somali, Hmong, and Latino communities, Jordan was intimidated by a 20-page application written at a university reading level. We were inspired to bridge this "Dignity Gap" by turning a bureaucratic interrogation into a simple, multilingual conversation. We realized that $30 billion in federal food aid goes unclaimed annually, not because of a lack of need, but because of a User Experience crisis.

What it does

BenefitFlow is a conversational AI "North Star" that helps users navigate the complex pre-application phase of SNAP.

Multilingual Pre-Screening: It conducts a 2-minute eligibility check in 28 languages (including Somali, Hmong, and Yoruba) using a 2nd-grade reading level.

Smart Logic: It applies 2026 MN-specific rules, including the new 80-hour work requirements for adults aged 18–64 and the 200% Gross Income limits.

Benefit Estimation: It provides users with an immediate estimate of their monthly benefits (e.g., up to $785 for a family of 3).

Application Readiness: It generates a personalized "Document Checklist" (ID, paystubs, utility bills) and directs users to MNbenefits.mn.gov or local navigators like HAP or CLUES.

How we built it

We engineered a production-ready stack designed for rapid deployment:

Frontend: Built with Next.js for a fast, mobile-first experience.

Conversational Engine: Powered by Vapi Voice AI, allowing users to speak naturally rather than typing into daunting form fields.

Logic & Policy: We hardcoded the 2026 Minnesota SNAP income tables into a custom LLM system prompt to prevent "hallucinations" of eligibility rules.

Communication: Integrated the Resend API to instantly deliver the personalized Document Checklist to the user's phone via email

Challenges we ran into

We got burnt out early on due to idea and features fatigue and trying to "perfect" our idea before starting out, we were discouraged but after doing more research we were able to bridge the gap of what general help looks live vs what it would look like in just Minnesota.

And the from there the primary challenge was "Technical Translation." Converting complex legal jargon like "Categorical Eligibility" or "Standard Shelter Deduction" into 2nd-grade level Somali or Hmong required intense prompt engineering.

Accomplishments that we're proud of

We are incredibly proud of our Linguistic Inclusivity. Seeing the AI seamlessly switch from English to Somali to Hmong—while maintaining perfect accuracy of Minnesota's gross income limits—proved that we could solve a massive barrier to food equity.

What we learned

We learned that in social services, "Scope is Kindness." We realized we didn't need to build a tool that filed the application, we needed to build a tool that empowered the person to file it. By focusing on Preparation and Estimation, we could reduce "Application Anxiety" and increase the success rate of the 20-minute official interview.

What's next for BenefitFlow

Our vision is to move from a "Pre-Screener" to an "Application Co-Pilot." 1. Live Handoff: Integrating a feature that connects a user directly to a live navigator at a Twin Cities non-profit if the AI detects high stress or complex immigration flags.

  1. SMS Version: Developing an SMS-only version for users without smartphones or reliable data plans.
  2. Scale: Partnering with Twin Cities food shelves to place QR codes on shelves, allowing residents to check their SNAP eligibility while they wait for emergency food.

Built With

Share this project:

Updates