MeNabung

Demo

About

The problem it solves **The Silent Crisis of Saving** In Indonesia, saving money the traditional way feels like fighting a losing battle. With inflation at **2.9%** and bank interest stuck at **0.4%**, keeping money in a bank account means you are literally paying for the privilege of losing value. We all know DeFi has the answer, yields of 5% to 15%, but actually *accessing* it is terrifying for normal people. I tried explaining "liquidity pools" and "impermanent loss" to my friends, and I saw them give up immediately. The existing tools don't help either; they are just chatbots giving generic textbook advice like "diversify your portfolio," without knowing if you have $10 or $10,000. **MeNabung bridges this gap.** We built a **"wallet-aware" AI** on Base. It doesn't just chat; it reads your on-chain **IDRX** balance, profiles your risk, and executes the strategy for you. It turns a complex, 10-step DeFi process into a simple, safe conversation, allowing everyday Indonesians to beat inflation without needing a degree in finance.

How it's made

Challenges we ran into ### The "Blind Advisor" Problem The biggest hurdle wasn't the smart contracts, it was the AI's context. In our first build, the AI was "blind." A user would ask, *"How much should I deposit?"* and the AI would give a generic Wikipedia-style answer like *"Financial experts recommend 20%..."* It was frustrating because it felt robotic. **The Fix:** We realized we couldn't just use a standard LLM call. We had to build a dynamic WalletContext system. We essentially "incepted" the system prompt with live on-chain data (User Balance: 978 IDRX, Current Vault APY: 12%) before the user even sent their first message. Suddenly, the AI went from a generic chatbot to a personalized advisor that said, *"You have 978 IDRX sitting idle; let's put that to work."* ### The IDRX Decimal Trap This one gave us a mini heart attack. We are used to working with ETH (18 decimals) or USDC (6 decimals). IDRX uses **2 decimals**. During our first integration test, the dashboard showed users having millions of dollars in savings because the math was off by a factor of 10^16. It was a mess. **The Fix:** We had to audit every single line of our frontend code. We implemented a strict utility function that checks the token contract's decimals *before* doing any formatting. No assumptions allowed. ### Making "Let's Go" Actually Work We noticed a funny UX issue: users would chat with the AI, get excited, and type *"Okay, let's do it!"*... and nothing would happen. The strategy card wouldn't pop up because our code was strictly looking for a *number* (e.g., "Deposit 500"). **The Fix:** We wrote a detectActionIntent() function. Now, if the AI detects strong intent keywords (like "ready," "let's go," or "invest"), it assumes the user wants to deposit their *current liquid balance* and triggers the transaction UI automatically. It made the experience feel 10x more magical.