6PM Finance
AI personal finance assistant that analyses your spending patterns and delivers plain-English insights every evening to help you take control of your money.
The Challenge
Translating raw transaction data from diverse banking institutions into genuinely useful, personalised financial insights — not generic tips — while maintaining the security standards required to handle sensitive financial data.
The Solution
Used Plaid for secure bank connectivity and built a transaction categorisation pipeline with Claude AI to generate contextual, narrative-style insights tailored to each user's actual spending behaviour.
6PM Finance is a personal finance assistant built around a simple premise: every evening at 6 PM, you receive a concise, conversational summary of your day’s spending and how it fits your monthly picture. By combining Plaid’s bank connectivity with Claude AI’s reasoning capabilities, 6PM turns the stream of cryptic transaction codes into a clear story about your financial habits — and what to do about them.
Challenge
Most budgeting apps display charts and category totals, then leave the interpretation to the user. The challenge was going one step further: generating insights that feel personally relevant rather than algorithmically generic. A transaction labelled “SQ *BLUESTONE LN” means nothing without context, and a pie chart showing “Food & Drink: 34%” doesn’t tell you whether that’s a problem. Building a pipeline that reliably enriches, categorises, and contextualises thousands of transactions per user required careful prompt engineering and a robust merchant normalisation layer.
Solution
The backend is a FastAPI application that pulls transactions via Plaid’s Transactions API, runs them through a merchant normalisation service to resolve cryptic codes to readable names, then batches them into a structured context window for Claude. A carefully engineered system prompt instructs Claude to produce insights in a consistent JSON schema — spending anomalies, recurring subscription alerts, goal progress, and a recommended action — which is then rendered into the evening digest notification. PostgreSQL stores enriched transaction history, enabling month-over-month trend analysis.
Key Features
- Secure bank connectivity via Plaid covering 12,000+ financial institutions
- Evening digest with AI-generated narrative spending summary via Claude API
- Automatic merchant normalisation for readable transaction descriptions
- Subscription tracker with renewal alerts and cancellation suggestions
- Goal-based savings tracking with projected completion dates
- Anomaly detection for unusual charges and potential fraud indicators
- Privacy-first architecture — raw credentials never stored, tokens encrypted at rest