Budget by Chatting: Building a multi-channel AI-powered expense tracker

Learn to build a multi-channel AI expense tracker. See how one LLM handles logging, questions, and feedback across SMS, WhatsApp, and web, with channel-aware responses and automatic feature request clustering.

Overview

SetForMoney is a household budgeting web app. Users text “groceries 45 Sams Club” to Telegram, WhatsApp, SMS, or a web chat and an LLM model parses the amount, matches the category, and detects intent from raw natural language.

Three things worth looking at under the hood:

The prompt architecture that handles expenses, questions, commands, and gibberish from a single parsing call, with fuzzy category matching that recovers from typos.

Channel-aware response formatting: same AI brain, different output constraints (SMS under 300 chars vs. rich web formatting).

Silent feature request detection: the chat assistant answers users normally while flagging unmet feature requests behind the scenes, then a nightly job clusters them into a demand-ranked product roadmap.

Live production demo.

Links

Tech stack