Inspiration
We were inspired by a staggering statistic: every year, over $100 billion in government benefits go unclaimed in the United States. This isn't because people don't need help, but because the gateway to public welfare—agency websites and paper forms—is built like a tax code. It is filled with legal jargon, rigid forms, and zero tolerance for error. For low-income single mothers, gig workers, and seniors under severe stress, this administrative friction acts as a systemic barrier. We wanted to build FormZero to humanize this interface and replace intimidating bureaucracy with a helpful, conversational companion.
What it does
FormZero is a privacy-first benefits navigator designed to unlock unclaimed government welfare. Instead of requiring users to navigate complex, intimidating official agency forms filled with legal jargon, FormZero provides a simple, conversational natural-language chat interface.
The application parses the user's household, income, and family situation, and scans eligibility rules across all 50 states using Retrieval-Augmented Generation (RAG). It then instantly generates a Matched Programs dashboard detailing eligible state and federal support (like SNAP, Medicaid, and Childcare subsidies) alongside:
Unclaimed Retroactive Benefits: Estimating the total back-pay they qualify to collect. Smart Deduplicated Document Checklist: An audit of all required files categorized by preparedness ("You Likely Have This" vs. Action Needed) that detects overlapping document rules and saves duplicate paperwork. Interactive Dependency Roadmap: A step-by-step application tracker that updates their annual value secured and bi-directionally checks off corresponding documents as they complete steps.
How we built it
FormZero is built as a split-architecture application:
Frontend: Next.js 16 with a high-fidelity glassmorphic CSS design system, bilingual translation (English/Spanish), local storage state persistence, and client-side calculators. Backend: FastAPI and Python, utilizing the Google Gemini 1.5 Flash API for natural language slot-filling and a FAISS vector database containing official policy rules for RAG context retrieval. To make eligibility concrete, we implemented real-time mathematical calculations. If a user qualifies for matched benefits with a combined monthly value of $V_m$, FormZero computes the unclaimed retroactive value $V_{retro}$ using a retroactive window of $M_{retro} = 41$ months (since January 2023): Vretro = Vm * Mretro We also track ongoing losses in real-time, calculating their loss rate per second $R_{loss}$ based on an average 30-day month: Rloss = Vm / 30*24*3600 In the Impact Analysis section, we compute the projected 3-year value $V_{3yr}$ and 5-year value $V_{5yr}$ to show long-term household impact: V5yr = 5*(Vm*12)
Challenges we ran into
50-State Abbreviation Collisions: In free-form text parsing, standard state abbreviations caused false positives (e.g. the word "in" matching Indiana or "ca" matching California inside words like "income" or "caseworker"). We solved this by implementing strict regex word-boundary checks. Bi-Directional State Syncing: Interconnecting the Document Checklist and the Dependency Roadmap was complex. When a user marks a program as applied in the roadmap, we had to dynamically update their unique document preparedness score, resolve overlapping file requirements (welfare programs sharing documents like identity/address proofs), and sync everything back. We resolved this by building a centralized React state calculator that deduplicates requirements in real-time. FPDF PDF-Generation Collisions: We ran into layout rendering exceptions where the FPDF2 library kept the cursor at the right margin when printing consecutive blocks of text. We solved this by creating a robust line-by-line parser that strips Markdown formatting and forces the cursor to reset to the left margin (pdf.set_x(10)) at the start of each iteration.
Accomplishments that we're proud of
Real-time Retroactive Benefit Clock: Building an interactive widget that calculates retroactive benefits and displays a ticking loss-per-second clock to create a sense of urgency. Smart Deduplication Engine: Designing a system that automatically identifies shared documents (like proof of ID or address) across different welfare programs so users don't have to gather or submit the same files multiple times. Privacy-First Local Caching: Ensuring that guest users can check their eligibility instantly while keeping all processing, calculations, and active tabs stored client-side for maximum privacy.
What we learned
We learned that the biggest barrier to social safety nets is cognitive load and administrative friction, not the criteria themselves. By translating separate agency guidelines into a singular, conversational workflow and deduplicating overlapping paperwork, we can reduce user stress and increase benefit capture rates. We also learned how to balance AI capabilities with strict responsible guidelines—using confidence scores and direct citation links rather than making binary assumptions.
What's next for FormZero: Privacy First AI Benefits Navigator
Pre-populated PDF Auto-Filling: Integrating PDF auto-filling libraries so that users can instantly download pre-populated official application forms. Localized 211 Community APIs: Connecting to local community APIs to list localized aid directories (e.g., food banks or local shelters) alongside state-level programs. Zero-Knowledge Proofs: Exploring zero-knowledge cryptography to allow secure document uploading and verification without exposing private data.
Log in or sign up for Devpost to join the conversation.