Nessential

Inspiration

Maximizing credit card rewards and loyalty programs is exhausting. People juggle multiple credit cards, restaurant apps, retail coupons, and confusing reward tiers—all while trying to stick to a realistic budget. Most budgeting apps are backward-looking (scolding you after you’ve already overspent), while deals apps urge you to spend money you don't have.

We built Nessential to create a truly "hands-off," proactive financial co-pilot. We wanted an assistant that understands what you can actually afford—factoring in rent, student loans, and even plans made in your group chats—and automatically stacks store loyalty perks with credit card rewards whenever you're nearby.

What it does

Nessential turns passive financial management into proactive, real-world savings:

Unified Financial Ledger & Onboarding: Seamlessly connects via Capital One's Nessie API to ingest checking balances, transactions, recurring bills, and loans into a unified financial profile.
Cashflow Forecasting & Routine Detection: Projects day-by-day cashflow and recognizes recurring habits (like Friday lunches or morning coffee runs) while reserving cash for bills and emergency funds.
Social Obligation Tracking via Texts & Email: Uses Photon to read incoming SMS messages and forwarded emails. If you agree to a $40 dinner split on Friday in a text, Nessential reserves that money before you can accidentally spend it.
Automated Loyalty Stacking: Automatically enrolls users into store loyalty programs (Chipotle, Panera, etc.) using private burner email aliases, stacking merchant perks on top of Capital One card rewards.
Location-Aware Smart Deals: When you are near a store, Nessential evaluates your hard budget and liquidity constraints, alerting you with the optimal purchase and reward recommendations via push notifications.
Conversational Voice Assistant: Users can pick up the phone and talk directly with an AI voice agent powered by Photon and Nemotron to check balances, review spending habits, or ask real-time affordability questions like "Can I afford to go out this weekend?"

How we built it

Mobile Client: Built with React Native and Expo (Expo Router, Location, Notifications, and Speech) for real-time GPS proximity scanning and mobile budget management.
Financial Core & Data: Integrated with Capital One’s Nessie API (v2) to read accounts, recurring bills, and loans, backed by a deterministic integer-cent forecast engine.
Language Intelligence: NVIDIA Nemotron (served via Brev / vLLM) utilizing structured JSON outputs to parse text message obligations, categorize caller questions, and explain recommendations without hallucinating financial math.
Voice, Text, & Email Pipelines: Powered by Photon for inbound/outbound interactive voice phone calls, incoming SMS text ingestion, and domain email forwarding for loyalty signups.
Persistence & Security: TigerData (PostgreSQL) for user profiles, memory, derived routines, and merchant loyalty catalogs with row-level security.
Constraint Optimization: A deterministic recommender and OR-Tools constraint solver to find the best reward stacking without violating strict user budgets.

Challenges we ran into

Separating Math from Language Models: Large language models are great at understanding conversational nuance but unreliable at arithmetic. We had to enforce a strict boundary: deterministic code handles all money, dates, and budget limits, while Nemotron handles natural language interpretation.
Multi-Modal Data Ingestion: Synchronizing asynchronous webhooks, incoming phone calls, SMS threads from group chats, forwarded emails, and GPS fixes without race conditions or double-counting obligations.
Loyalty Program Realities: Navigating merchant signup flows, email verification codes, and privacy considerations without fabricating identity data or bypassing bot defenses.
Unifying Divergent Microservices: Integrating independent services across Node.js, Python, PostgreSQL, and Expo into an automated, end-to-end data pipeline.

Accomplishments that we're proud of

A True End-to-End Autonomous Pipeline: Connected the entire loop—from reading a bank account and scanning group texts to automated loyalty enrollment and live GPS deal notifications.
Conversational Voice Agent with Guardrails: Built an interactive voice assistant that handles complex questions like "Can I afford dinner at Pappas on Friday?" grounded strictly in real balances, upcoming bills, and pending text splits.
Zero Hallucination Budgets: Guaranteed that budgets are treated as mathematically uncompromisable constraints—Nessential will never recommend a purchase that jeopardizes financial health.
Resilient Multi-Channel Testing: Achieved extensive automated test coverage across mobile bundle builds, backend ledger migrations, voice adapters, and solver optimizations.

What we learned

AI Needs Strong Deterministic Anchors: In fintech, user trust hinges on accuracy. Using structured JSON schemas with rigid code-level validation is essential when pairing LLMs with financial data.
Conversational Interfaces Make Finance Accessible: Users are far more likely to engage with their finances when they can talk naturally over a phone call or get a timely notification rather than clicking through complex dashboards.
System Architecture Under Pressure: How to cleanly structure and coordinate distributed services (AI VMs, voice telephony, database migrations, and mobile apps) in a fast-paced hackathon environment.

What's next for Nessential

Expanded Merchant Partnerships: Automating loyalty onboarding across a wider catalog of national dining and retail programs.
Background Geofencing & Native Push: Transitioning from foreground polling to power-efficient native background geofences and APNs/FCM push notifications.
Group & Split Integration: Directly requesting and settling peer-to-peer payments when text obligations are detected.
Point-of-Sale Integration: Exploring Apple Wallet / Google Pay integration to automatically surface the highest-yielding card and scanned loyalty barcode right at the checkout terminal.