RASEED is an AI-powered receipt and financial assistant integrated with Google Wallet. It turns everyday transactions into structured, queryable records and actionable financial guidance — receipt ingestion, budgeting, and spending insights, plus recurring-payment tracking, family mode, recipe generation, and tax assistance, all powered by Gemini models and Vertex AI.
Receipts come in three ways: photograph or upload them and Gemini extracts the metadata, dictate the purchase by voice, or type it directly into the entry form. A cloud function then classifies each item as recurring, warranty, or food based on database triggers, routing it into the right table automatically.
Confirmed receipts can be pushed straight to Google Wallet as a custom pass — built on generic-pass classes per receipt category, with hero images and color coding, text and link modules carrying metadata on the back of the pass, and a deep link back into the app.
Recurring items get tracked against their own deadlines, with reminders sent as a payment approaches; warranty items are tracked the same way, either parsed straight from the receipt or added manually against a specific item.
Spending is broken down into categories — Groceries, Entertainment, and the rest — with per-category budget limits, over-budget flags, and a custom date range for reviewing any window of activity. Family mode extends the same view across a household: add members by email and see joint analytics across the same curated categories.
Agents
A recipe generator agent proposes recipes from the food items a user has actually bought, respecting whatever dietary restrictions or time constraints are given. It pulls from a database-access tool and a Google Search tool, merges the two through a combining agent, and runs the result through a validator agent that checks instructions were followed and keeps hallucination in check.
A tax and reimbursement agent does two things: a Vertex AI tool classifies every receipt into a fixed set of categories and totals the indirect tax paid per category, while a Firestore tool surfaces which receipts are eligible for reimbursement based on their category and subcategory.
Future scope
Planned next: ML-driven spending analysis that predicts expenditure patterns and nudges budgets before a category runs over — down to context-aware suggestions like recommending a bus over a cab when a travel allowance is running low. Google Pay auto-ingestion, registering RASEED as a share-target app so a shared GPay receipt gets parsed and tagged with zero manual entry. And context-aware purchase recommendations, where a chatbot query like "I want to buy paneer tikka" returns ranked options across price, proximity, offers, and sustainability.