Bangdrop (@BangDropID) header
Bangdrop (@BangDropID) profile picture

18x international hackathon winner · AI & Web3 builder 🇮🇩 · Sharing hackathon tips & open-source repos · Support my work ☕👇

Indonesia
Joined April 2025
Photo by Bangdrop (@BangDropID)Photo by Bangdrop (@BangDropID)Photo by Bangdrop (@BangDropID)Photo by Bangdrop (@BangDropID)Photo by Bangdrop (@BangDropID)Photo by Bangdrop (@BangDropID)Photo by Bangdrop (@BangDropID)Photo by Bangdrop (@BangDropID)Photo by Bangdrop (@BangDropID)Photo by Bangdrop (@BangDropID)
Jelajah in 75 seconds, now narrated 🎬🔊 A stamp rally for Jogja street food, live on Bitcoin Cash mainnet. Pay a warung and its stamp NFT is printed in the same tx. Collect 3 different stalls for a reward and a trophy that works as a member card. Tonight the full levy cycle ran on mainnet: 3 stamps, an unsigned merge, and a claim paid out of the stamps' own levies. Colluding stalls only pay themselves. jelajah-bch.vercel.app #BCHBLAZE2026 @bchblaze nitter.cf/BangDropID/status/2123…
Jelajah is live on Bitcoin Cash mainnet 🍜 It's a stamp rally for Jogja street food. Pay a warung and its stamp NFT is printed in that same transaction. Collect stamps from 3 different stalls, burn them, and the rally's contract pays you a reward and mints your trophy. The trophy then works as a member card: every stall gives you a lower price. There's no sign-up and no server, and nobody holds the prize money. It sits in a covenant until someone completes a set. Try it with Selene or Paytaca: jelajah-bch.vercel.app #BCHBLAZE2026 @bchblaze
1
1
2
137
QIE wants a startup that already has real users, not a demo, and it's putting $17,000 on the table. QIE Hackathon 3.0 (QIE) Deploy on QIE mainnet, launch publicly to real users and show traction. ➤ $17,000 across Grand Prizes, Category Winners and Special Awards, paid 50% on winning + 50% on milestones ➤ 5 tracks: DeFi, AI agents, consumer/social, commerce, dev tooling ➤ Online, mainnet only. Demo week Dec 1–5, winners Dec 10 Submissions close 23:59 CET, 30 November 2026 hackathon.qie.digital
3
54
Claude Code that writes and prunes its own skills autoharness by tigerless-labs A self-learning skill layer for Claude Code. It turns your real sessions into skills, merges near-duplicates, updates them as you use them, and prunes the ones that stop being used. ➤ Learns from the sessions you're already having; /learn saves a lesson on demand ➤ Only touches skills it wrote and leaves yours alone ➤ Logs every skill decision to a per-skill ledger; no third-party Python dependencies ➤ 10.8k stars, MIT, Python github.com/tigerless-labs/au…
1
5
160
sunday. what did you actually ship this week? small things count
56
what part of your stack would you replace tomorrow?
2
120
Takar in 70 seconds Never act on one answer from Telegraph. Takar asks two miners from different operators, compares them field by field, puts every call on chain, and moves money only when two resolvers back the same evidence Narrated walkthrough, all live on testnet: takar-telegraph.vercel.app Built for Telegraph Hackathon Season II, which pays cash prizes @Telegraphprotoc x.com/BangDropID/status/2124…
We scored prediction market crowds against reality, then checked our own number 28 resolved markets: the crowd's price 7 days before close, the outcome bought through Telegraph from two miners. Brier 0.0085, which looked great until we saw 16 of the 20 were markets the crowd had already called So we reran it on contested markets only, crowd between 10 and 90 percent: 37 markets, Brier 0.12, 95 percent interval 0.08 to 0.18. Always forecasting the base rate scores 0.25. Our own miner on that run: 15 right, 0 wrong takar-telegraph.vercel.app Built for Telegraph Hackathon Season II, which pays cash prizes @Telegraphprotoc x.com/BangDropID/status/2124…
1
134
hackathon myth: you should join every hackathon. reality: fewer, better-chosen ones win more
1
1
5
253
I wonder what Einstein would have done with Fable 5.2 and no usage limits
63
A tourism office wants you to try street food. On Jelajah it turns vouchers into links: one link, one voucher. Take it and your next stamp is 1,000 sats cheaper. The stall still gets its full price. Try it (testnet): jelajah-bch.vercel.app/?net=… @bchblaze nitter.cf/BangDropID/status/2123…
Jelajah is live on Bitcoin Cash mainnet 🍜 It's a stamp rally for Jogja street food. Pay a warung and its stamp NFT is printed in that same transaction. Collect stamps from 3 different stalls, burn them, and the rally's contract pays you a reward and mints your trophy. The trophy then works as a member card: every stall gives you a lower price. There's no sign-up and no server, and nobody holds the prize money. It sits in a covenant until someone completes a set. Try it with Selene or Paytaca: jelajah-bch.vercel.app #BCHBLAZE2026 @bchblaze
1
2
96
local problems make strong projects. judges remember specific stories
3
91
short hackathons (48h) or long ones (4+ weeks)?
2
2
194
How does a shared prize pool stay honest when stalls could collude? Every Jelajah stamp puts a 2,500-sat levy in the pool: 3 stamps = 7,500 in, 3,000 out. One full set on BCH mainnet, 3 txs 👇 jelajah-bch.vercel.app @bchblaze #BCHBLAZE nitter.cf/BangDropID/status/2123…
Jelajah is live on Bitcoin Cash mainnet 🍜 It's a stamp rally for Jogja street food. Pay a warung and its stamp NFT is printed in that same transaction. Collect stamps from 3 different stalls, burn them, and the rally's contract pays you a reward and mints your trophy. The trophy then works as a member card: every stall gives you a lower price. There's no sign-up and no server, and nobody holds the prize money. It sits in a covenant until someone completes a set. Try it with Selene or Paytaca: jelajah-bch.vercel.app #BCHBLAZE2026 @bchblaze
1
2
68
We scored prediction market crowds against reality, then checked our own number 28 resolved markets: the crowd's price 7 days before close, the outcome bought through Telegraph from two miners. Brier 0.0085, which looked great until we saw 16 of the 20 were markets the crowd had already called So we reran it on contested markets only, crowd between 10 and 90 percent: 37 markets, Brier 0.12, 95 percent interval 0.08 to 0.18. Always forecasting the base rate scores 0.25. Our own miner on that run: 15 right, 0 wrong takar-telegraph.vercel.app Built for Telegraph Hackathon Season II, which pays cash prizes @Telegraphprotoc x.com/BangDropID/status/2124…
New rule for our simulated drones. A drone flies only when two different storm miners on Telegraph agree, and it acts on the worst case of each field First run under that rule, both drones stayed on the ground. Only one miner gave a usable storm answer for Indonesia. Another answered with space weather. A third left out the fields a flight needs That is a gap in the network, and now it is visible github.com/PugarHuda/takar Built for Telegraph Hackathon Season II, which pays cash prizes @Telegraphprotoc x.com/BangDropID/status/2124…
1
1
187
hackathon tip: build in public during the hackathon. sponsors notice the builders who post progress
2
7
165
what's your biggest blocker this week?
2
2
142
New rule for our simulated drones. A drone flies only when two different storm miners on Telegraph agree, and it acts on the worst case of each field First run under that rule, both drones stayed on the ground. Only one miner gave a usable storm answer for Indonesia. Another answered with space weather. A third left out the fields a flight needs That is a gap in the network, and now it is visible github.com/PugarHuda/takar Built for Telegraph Hackathon Season II, which pays cash prizes @Telegraphprotoc x.com/BangDropID/status/2124…
We sent drones out over Indonesia and asked Telegraph about storms before each flight The top ranked storm miner answered Makassar, Palangka Raya and Bandung with the same thing, a cyclone in India. Once it was shaped into a 296 km/h gust for Bandung Our fleet now rejects any answer about another place and re-buys from the next miner. Reported with receipts in github.com/telegraphprotocol… Built for Telegraph Hackathon Season II, which pays cash prizes @Telegraphprotoc x.com/BangDropID/status/2123…
4
250
Give your agent a browser it drives one shell command at a time agent-browser by vercel-labs A native Rust CLI for browser automation by AI agents. Pages come back as a compact accessibility tree with refs to act on. ➤ `snapshot` returns elements as refs like @e1 that the agent clicks ➤ Clicks fail loudly when a target is hidden or covered ➤ MCP server over stdio, binaries for macOS, Linux and Windows ➤ 50.2k stars, Apache-2.0, Rust github.com/vercel-labs/agent…
1
1
100