DeFi Monitoring Agent · 2025
Role
Design + Build
Year
2026
Platform
Web + Telegram
Type
Live agent
Overview
A vault monitor that explains what changed and why it matters
Lido Vault Alert Agent helps EarnETH and EarnUSD users stay aware without reading contracts, dashboards, and API responses all day. I designed the alert experience around plain language and custom rules for severity, frequency, and summary delivery. Then with the help of my ai agent built the live Next.js app, Telegram bot, API endpoints, and subscriber delivery logic. This project was for the Synthesis Hackathon a web3 hackathon that happened March 2026.
The Problem
Vault users need signals, not another dashboard to babysit
Earn vault positions can change because of APY shifts, TVL movement, curator rebalancing, benchmark underperformance, or wallet level position changes. The UX challenge was turning technical monitoring into timely alerts people can understand and act on.
Hidden changes
Vault APY, TVL, allocations, and wallet positions move across different sources that are not displayed to vault depositors.
Alert fatigue
Not every change deserves an interruption, so the system needed severity, frequency, and summary delivery rules.
Technical language
A useful output is not raw data; it is a readable explanation of what changed and whether action is needed on the depositors part.
Process & Exploration
Designing alerts around decisions, not data dumps
I mapped the monitoring journey from vault signal to human-readable notification: what changed, how severe it is, who receives it, and whether it should interrupt immediately or wait for a summary. That structure shaped both the bot commands and the API response design.
Notification rules
Critical alerts send immediately; lower-priority changes are bundled into morning and evening digests.
Signal model
APY drops, benchmark gaps, TVL changes, allocation shifts, and wallet position reads become alert candidates.
Bot UX
Telegram commands let users subscribe, set sensitivity, add email, check status, and unsubscribe.
Design Solution
Three layers that made the agent useful
01
Live vault reads
The app pulls APY and TVL from Mellow, benchmarks from Lido and DeFiLlama, and wallet/vault data from on-chain reads.
Plain-language alerts
The alert engine converts technical changes into severity, event type, explanation, and suggested action so users are not left interpreting raw numbers.
Agent-ready API
The same data is exposed through JSON endpoints, so bots, agents, and dashboards can ask for health, alerts, and yield-floor status.
The Build
Designed the alert system then shipped the live bot
My AI agent handled the heavier technical lifts in Next.js and TypeScript, I set up Supabase for subscribers, Telegram delivery, Vercel cron routes, and live Ethereum reads leaning on the agent whenever I needed help. The project went beyond a UI concept: it shipped as a working agent with a live bot, health endpoints, and delivery configuration.
Next.js
TypeScript
Telegram Bot API
Supabase
Vercel Cron
Lido SDK
On-chain reads
API + bot
$ GET /api/health?wallet=0x…
Telegram subscribers get critical alerts immediately and lower-priority changes bundled into digests.
Outcome & Reflection
What shipped and what I learned
3
Public API Endpoints
2
vaults monitored
Live
Telegram + email path
This project helped me connect UI/UX thinking with real-time system design. The hard part wasn't showing more data it was deciding which data should become an alert, which alerts deserve attention, and how to phrase them so a depositor can quickly decide whether to act. I also picked up new tools along the way: Vercel cron routes, live Ethereum reads, and Alchemy (a blockchain developer platform with infra and APIs for reading on-chain data). Next I'd add richer alert history, wallet connect, and a visual timeline of vault events.
Final outcome