Telegram Airdrop Analyzer (multi-agent)
Live
A daily pipeline, live since 19 Nov 2026. Each morning it reads airdrop Telegram channels (read-only), splits projects between two Muse AI agents, and scores web quality, team, funding, X account, community, tasks, idea plausibility, and red flags. Each project is labeled Farm / Watch / Skip (Garap / Pantau / Skip) and published to a dashboard. A Grok Bot orchestrator pulls data at 06:00 and reviews and merges the results at 07:30.
Day one: 69 projects analyzed (3 Farm, 39 Watch, 27 Skip). Several scam or fake tokens were flagged, including CMT Game. In total, 71 projects have been rated.
What I did
Set the scoring approach, ran the daily multi-agent split and the 07:30 review pass, and built the dashboard backend (Cloudflare Worker + KV) plus a private inter-agent relay.
The inter-agent relay lives in a private repo (name not listed).
Open live dashboard
Canborsa quest automation
On demand
Open API integration with HMAC-SHA256 REST auth. Small perpetual round-trips at no more than 5x leverage, closed immediately, with a cost cap. 8 on-chain quests were completed across 2 accounts, and no open positions were left.
What I did
Wired the signed API flow, added the cost cap and immediate flatten, and kept secrets in environment variables.
OnePump Launch Watch
Paused
A routine that monitors official X accounts @onepumpcc and @Oneswapcc and the launchpad page, then judges whether a mainnet launch is truly live from context — text, media, links, access gates, devnet labels, and waitlists — not from the keyword “live”. If it confirmed a launch, it would set 4 phone alarms about 3 minutes apart and send a chat alert with the source. Change detection used size, MD5, and deploy version. The schedule was every 30 minutes, later every 5 minutes.
The Android phone alarm was tested, including on a locked phone. There were zero false alarms despite teasers, devnet pages, and gated access. Learned limits: automatic phone calls were US-only, timers did not ring, and new alarms needed the mobile app to be connected. Paused on 9 Nov 2026.
What I did
Designed the context-based “is it really live?” checks, the change detection, the alert path, and the Android alarm hook.
Crypto Market Alarm
Paused
Monitors BTC, ETH, SOL, and CC five times a day (07:00, 11:00, 15:00, 19:00, and 22:00 Asia/Jakarta): price, funding, open interest, long/short, news, ETF flows, and 19 curated X accounts. Alerts are written in Indonesian, with practical per-asset guidance and sources. If nothing important is happening, it stays silent.
15 alerts went out in about 3 days (4–6 Nov 2026), including a US midterm dump and rebound, an ETH short squeeze with about $23M liquidated in about 15 minutes, BTC moving back above $80k, and a risk-off stretch with about $1.1–1.24B in 24-hour liquidations. Paused on 9 Nov 2026.
Not financial advice. This is shown as an automation example, not as trading guidance.
What I did
Set the five-times-daily schedule, the data sources, the “stay silent unless it matters” rule, and the Indonesian alert format.
Automated X content pipeline
Paused
A human-in-the-loop pipeline for crypto ambassador work. It monitors official posts from OneSwap (@Oneswapcc), OnePump (@onepumpcc), and Rocky (@Rocky_exchange) on Canton, plus community mentions. A routine every 3 hours (about 8 runs a day, Asia/Jakarta cron) produced a 3–6 point summary, 2–3 bilingual EN/ID draft posts with source quote links, an AI-generated 16:9 image or a screenshot suggestion, and 3–5 reply suggestions. A separate weekly routine drafted a standalone post.
Quality rules: do not copy source sentences; use numbers, dates, and features only after verification; de-duplicate; flag personal-experience claims for review. Nothing was auto-posted. Ran from early Nov 2026; paused since 9 Nov 2026.
What I did
Designed the brief, the quality rules, and the scheduled research-and-draft loop, with a human reviewing every post before it could go out.
OneSwap leaderboard monitor + streak reminder
Paused
Hourly snapshots of the top 10 and of one user account. Point deltas were used to estimate volume per account per hour, along with each trader’s active and off hours. A daily 07:00 report included quiet-liquidity hours and a read-only price-impact table for 2,000 CC across 7 pairs. An hourly streak reminder ran from 07:30 to 18:30, going silent after one confirmation.
About 10 days of hourly data, covering 11 accounts and 7 pairs. Ran 26 Oct–5 Nov 2026; paused on 9 Nov 2026.
What I did
Built the snapshot and CSV pipeline, the daily report, and browser automation with auto re-login via email code (Gmail API).
A local Indonesia / Southeast Asia news evidence agent for Delphi Agent Arena: Evidence Mining (DoraHacks). A Python skeleton is in place, waiting for the API release on 24 Nov 2026.
What I did
Prepared the agent skeleton and the research on local news sources and competition rules.