PesaMind Agent

• 

  AWS DynamoDB that stores the Category mappings

• 

  WhatsApp Category Request Prompt

Inspiration

Keeping track of daily expenses is exhausting. Mobile money receipts (like M-Pesa) drop into our phones all day, but opening a budgeting app to manually type in every single meal, bus fare, or grocery run takes too much effort. Most people give up on budgeting simply because manual entry is a chore. That is why I built PesaMind; a smart assistant on WhatsApp that turns raw text messages into organized spending insights automatically, so you can track your money without doing any work.

What it does

PesaMind handles your daily spending tracking from start to finish: • Reads your receipts: Converts incoming text messages into organized payment records instantly. • Sorts your spending: Automatically groups your money into clear categories like Food & Groceries, Utilities, or Transport. • Asks when unsure: If it sees a shop or name it doesn't recognize, it sends you a quick WhatsApp message with a list of options so you can pick the right category. • Remembers your choices: The next time you pay that same shop, PesaMind remembers your choice and categorizes it automatically. • Answers your questions: Text it like a friend on WhatsApp (e.g., "How much did I spend this week?" or "Show my food spending") to get instant totals.

How we built it

PesaMind brings together four main parts:

1. The Automation: A phone app forwards incoming payment text messages directly to the cloud.
2. The AI Brain: Built using Amazon Bedrock (Claude Haiku) and the Strands framework, which acts as the intelligent assistant capable of reading texts, picking the right tools, and answering questions.
3. The Database: Amazon DynamoDB safely stores your spending history and remembers your merchant rules.
4. The WhatsApp Interface: Connected via Twilio, allowing PesaMind to send you alerts and chat with you directly inside WhatsApp.

Challenges we ran into

Early on, freeform text responses with special characters (like & or <) caused WhatsApp delivery to fail silently. I fixed this by wrapping the assistant's responses in safe web formatting tags, ensuring every message reaches your phone reliably. Another challenge was stopping simple greetings (like "hi" or "help") from triggering database searches. I added dynamic routing to the AI prompt so it instantly recognizes greetings and replies with a friendly menu, keeping the system fast and responsive.

Accomplishments that we're proud of

• A Chatbot That Learns: Built a true interactive assistant that asks for your input when stuck and remembers your answer forever. • Zero Cost When Idle: The cloud infrastructure only runs when a message comes in, keeping costs at $0.00 when you aren't using it. • Natural Money Questions: You can ask about your spending for today, this week, or specific dates using normal conversational language.

What we learned

I learned that the best way to make AI reliable is to let it handle the language understanding, while traditional code handles the database and math. Using the Strands framework allowed the AI brain to invoke tools smoothly while letting deterministic code handle exact mathematical calculations.

What's next for PesaMind Agent

Setting up custom budget warnings when spending is too high, and sending weekly spending charts right to your WhatsApp chat.

Built With

• amazon-bedrock
• amazon-sdk
• api-gateway
• aws-lambda