🚨 BANG DROP ZONE IS OPEN. Not for lurkers. For people who BUILD. Hackathons • AI • Web3 • Dev Alpha • Resources • BTS • Opportunities And I decided to make this group public but with more strict rules ⚠️ RULES • 3 days inactive → auto kick • No spam / random promo • No drama / toxicity • Keep shared info private • Contribute, don’t just watch • Quality > quantity Free to join. No paid access. Show up. Build. Level up. 🚀 🔗 t.me/BangDropZone Build. Share. Vibe.

WE PLACED 🥉 3rd place in Track 2 (Evaluation WASMs) at the @Telegraphprotoc Hackathon Season I! $200 🎉 Plot twist: I spent most of the month building a storm-risk miner and an app on top of Telegraph. The thing that actually won was the scorer 😂 amanat_scorer is 1,411 lines of no_std Rust squeezed into ~16 KB of WASM. It doesn't read answers as text. It pulls out the numbers and grades them as real measurements: 🌡️ 38.2 °C = 100.8 F = 311.35 K, all the same reading ✅ 0.3° off → right ❌ 30° off → wrong And it got knocked down A LOT on the way: → First WEATHER_FORECAST registration rejected: tied 15–15 on wins, but a 0.82 margin vs the champion's 0.99 → Took the GAME_RESULT champion slot… and held it for about 40 minutes 💀 Every rejection came back with the exact numbers, so I always knew what to fix next. That's what makes Telegraph fun: the evaluation layer is open, anyone can challenge it, and you can see exactly why you lost. Huge congrats to @zkasuran and @hyadav42774, and to every winner across all three tracks 🙌 Season II: $10,000, 30 days. I'm coming back for more 🔥 github.com/PugarHuda/amanat/… #TelegraphHackathon

Gadai — a credit desk for @bankrbot agents. An agent earns a cut of every trade of its token, but the fees land later and the compute bill is due now. Its options: top up by hand or sell. The lien. The borrower points its Doppler fee share at a per-loan FeeVault with updateBeneficiary. Only the contract moves it back, only to the borrower, and release() is permissionless; the desk can't keep them. The funding. Each loan mints a FeeNote, an ERC-20 whose face value is the debt, sold for USDC in a @Uniswap Continuous Clearing Auction. Collected fees repay it via the Uniswap Trading API. The underwriter reads real Bankr fee history and writes three memos; the lead is binding and can only lower it. Before approving, its @dynamic_xyz MPC wallet pays $0.05 USDC over x402 for a honeypot verdict, and uses it. Every call is a public signal you can mirror with @DefinitiveFi Flash: an entry plus a Bracket (TP/SL), or a DCA from a TWAP. The desk went first: TWAP fb3b2572-48c6-4ce5-b79a-6199cefac82f, filled on Base mainnet. The Credit Line Board prices every Bankr agent: 105 on Base, 14 eligible, $856.57 pre-approved ( UTC, 20 Sep). Three are paid in tokenized stocks, so the desk bought one: 0.00079 TSLA on Robinhood Chain via Relay + Uniswap. A Blackbird Flynet concierge plans a meal inside the dining budget. ERC-8004 agent #94699 carries repayment reputation; @base builder code bc_32d4pc8g on every desk tx. Honest scope: the full lifecycle ran on an Anvil fork of Base against a real Bankr pool. The mainnet actions are at /evidence. gadai-six.vercel.app github.com/PugarHuda/gadai

AND THIS IS MY 12th WIN HACKATHONNNN i still didn't expect that i won in this super big event from eth

WE WON. 🥈🔥🔥🔥 Batas just took 2nd place in @ensdomains Best Use of ENSv2 at @ETHGlobal #ETHOnline2026 and I'm honestly still shaking 😭🙌 I built this solo. No team, no existing codebase, just a blank repo, an idea, and way too much coffee ☕️ There were so many moments where it felt like it wasn't going to come together. Tests breaking, deploys failing, things that worked yesterday suddenly not working today. I kept going anyway. And now here we are. 2nd place. 🥈 Seeing Batas on that winners list hit different. It's proof that one person with a stubborn idea can still ship something the judges notice ✨ Huge thank you to the ENS team for building something this fun to build on, and to the judges for taking the time to actually dig into it 💙 Thank you @ETHGlobal for running an event that pushes you way past what you think you can do in a few days 🙏 And thank you to everyone who checked out the project, dropped a reply, or cheered me on along the way. It meant more than you know ❤️ This is just the start. Batas isn't done yet 🚀 👉 batas-one.vercel.app

Attestcoin can prove a transaction happened on Ethereum. It can't prove a list of transactions is complete. Whoever submits the proofs picks which ones to submit, and every one of them verifies. So "this borrower has never been liquidated" is a sentence no inclusion proof can ever back. That's the sentence every on-chain credit product needs. I built Utuh for BUIDL CTC on @Creditcoin Hackathon to close that gap. A claimant bonds the claim that a set of events is whole. Every member is verified by the Block Prover on the way in. If anyone finds one in-scope event the claim left out, one proof breaks it and pays them half the bond. Presence stays cryptographic. Absence becomes economic. On top of it, a credit line in CTC against real Ethereum history. Nothing bridges, only proof crosses. What's live right now: • 34 claims refuted on-chain, 34 CTC burned • A false "never liquidated" claim over 216,000 blocks of Ethereum mainnet, broken by one liquidation proof • A browser console that sweeps Ethereum and refutes with no server, no key, no wallet needed to look • The watcher role as an MCP server, so an AI agent can hold it: npx -y utuh-mcp • 193 tests, and one command that re-checks every number above against the chain Live app: utuh.vercel.app See a claim somebody broke: utuh.vercel.app/?deployment=… Demo: youtu.be/HwSnv3E4tzo Code: github.com/PugarHuda/utuh @DoraHacks

Just submitted Batas to @ETHGlobal #ETHOnline2026 🛑 An AI agent runs your @1inch Aqua liquidity. The contract holds the limits. Batas is Indonesian for "mandate": authority within limits that must not be exceeded. The problem Aqua's pull() only checks who is calling. Whatever app you ship liquidity to can move your tokens. Hand that app to an autonomous agent, and nothing on-chain stops it. How Batas stops it • The mandate IS the strategy. Cap, floor and expiry are encoded into the Aqua strategy bytes, so dropping a limit gives a hash nobody funded • Two new SwapVM opcodes enforce it inside settlement: PolicyEnvelope (0x21) and MandateName (0x22) • A two-sided banded position, shipped live on Sepolia Kill switch on @ensdomains ENSv2 The agent's authority is agent.batas.eth: expiring, soulbound, revocable. MandateName reads it on every settlement. Revoke the name and every swap reverts, for every caller. Plus a PermissionedResolver, wildcard, aliasing, delegated text records, and a two-way ENSIP-25 link to ERC-8004. Agent payments on @hedera • x402 via Blocky402, priced per request by the work it takes • pay in HBAR or our HTS token with a custom fee • every payment audited on HCS • agents negotiate a fill over A2A and settle in the same task • recurring payments with Scheduled Transactions • discoverable through ERC-8004 and the Hashgraph Online directory Everything runs live on Sepolia and Hedera testnet, and the test suite checks the README's claims against the chain. Built solo, from scratch. Live: batas-one.vercel.app Code: github.com/PugarHuda/batas

Don’t forget claim ur reward guys from private sprint strk20 hackathon

agent memory stored as markdown files sqlite-memory does semantic search and hybrid retrieval over plain markdown, offline-first, with sync between agents the format choice is the whole thing. your agent's memory stays readable by you, greppable, and diffable in git. every other option in this space hands you an opaque store you can only query through their API github.com/sqliteai/sqlite-m…

Everyone says their bot adds liquidity. I measured mine. Every window it ever quoted on @dreamDEXSomnia 148 of them rebuilt twice from @Somnia_Network's own order rows: once with my quotes in the book, once with them removed. Spread 0.0245 without. 0.0175 with. 28% tighter,The timeline cuts about here everything below is behind “Show more” 138 improved, 0 worse. That is not a backtest. It is the venue's own Order rows replayed at the instant each quote landed, and it is the one claim in this hackathon I have not seen anyone else make. The mechanism is why it can quote both sides holding nothing. On DreamDEX two opposite-side BUYS cross with no seller at all the pool mints a fresh Up/Down pair out of their combined collateral. Hold one of each and it redeems for exactly 1 whichever way the window resolves, and it cost less than 1. The spread is collected, never bet. Directional exposure: 0.00. And the vault is down 5.80%. -989.83 realised on 28,994.67 of basis. 25% of filled quotes went adverse, where the strategy needs under 9%. One leg fills, the market walks away from the other, and NAV marks that naked leg at zero so the loss lands on whoever is already in rather than on the next depositor. That number is on the front page, not in a footnote. A record you only publish when it is green is not a record. Two things went back to the venue instead of into my repo. Sixteen reproducible SDK and venue defects, filed on their own tracker with transaction hashes. Six are now a section of their documentation a Somnia collaborator re-measured every entry against mainnet and Shannon, corrected three of mine against newer contract behaviour, and landed it: "All six reproduce... The section is yours." And 637 of the venue's function and error signatures published to 4byte, so a DreamDEX revert decodes in cast 4byte and Blockscout for every team on the chain, not just for me. Unprompted, from the same collaborator: "The 4byte registration is real and already earning its keep: OnlyApprovedContracts() is there from 6 Sep and decoded a revert for me in an unrelated investigation this week." Their words, not mine. That is the only part of an ecosystem claim that is not self-reported. Built solo. 149 unit tests, five stateful invariants, nine fork tests against the real venue, 103 browser tests, a16z's ERC-4626 property suite. Every number above reads from the chain in your own browser. abadi-wheat.vercel.app github.com/PugarHuda/abadi

Build an autonomous agent, run it live for 14 days, get audited Beyond the Benchmark: The Binance Agentic AI Challenge 2-week build, 2-week scored run. You build, host and operate it. Binance sets rules and verifies the record. No personal capital, ever. ➤ $6,000 in prize assets across top 3 ➤ Binance Seeds internship and full-time opportunities ➤ Under 25 finalist teams from open applications ➤ Teams of 1 to 4 ➤ Scored on wallet equity change, tie-break on lower max drawdown ➤ No live leaderboard, full 14 days count Applications Sep 9-30. Only 10 hackers registered so far. dorahacks.io/hackathon/binan…

I’m indonesian but my audience mostly from nigeria ahaha love to engage with nigerians And need work more to get attention audiences my own country

Integrate KeeperHub into a live project, $5,000 in stablecoins KeeperHub: The Agent Economy Hackathon @KeeperHubApp One main track plus a bounty. Every submission reviewed at repository level. ➤ Main track $4,000, ranked: $2,000 / $1,200 / $800 ➤ Bounty $1,000, two winners at $500, ship a merged PR to the KeeperHub repo ➤ Bounty stacks with the main track ➤ Required: source link, demo video, and a link to a transaction executed through KeeperHub ➤ Top 10 present live to judges Build phase Sep 6 to 18. 171 hackers registered. dorahacks.io/hackathon/keepe…

microsoft archived typescript-go on september 1, and the archive is the announcement it was the staging repo for rewriting the TypeScript compiler in Go. that port is done. it shipped as TypeScript 7.0 and moved back into the main repo during the preview you ran it as `tsgo`. from the 7.0 RC onward the command is just `tsc` again a rewrite of a language toolchain that ends with the command name unchanged is the rarest possible outcome. almost none of them land github.com/microsoft/typescr…

Build anything on the CoinMarketCap API, three weeks Build with CMC: API Hackathon @CoinMarketCap Free Startup tier for every participant. Prices, market caps, exchange volumes, DEX pairs, derivatives, on-chain and the new RWA endpoints. ➤ Top 3 get a full year of CMC API access after the event ➤ Top 30 get a swag box ➤ Notable projects retweeted by CMC, 7.1M followers ➤ Four tracks: trading tools, AI agents, data viz, RWAs ➤ Scored on does it work 30, usefulness 25, API use 20, code quality 15, presentation 10 Submissions close Sep 30, judging to Oct 16 dorahacks.io/hackathon/coinm…

an Anthropic pretraining researcher quit the AI industry today, and the timeline is the part worth sitting with Jacob Coxon, 27, British mathematician. three years of pretraining research across OpenAI and Anthropic. he was at OpenAI from 2023 until July 2026, then moved to Anthropic specifically because of its safety reputation. two months later he resigned and left the field entirely his charge is that neither lab is acting responsibly, and that both are racing straight to self-improving superintelligence and gambling with our lives the distinction he drew between them is the uncomfortable part. at OpenAI, he says many haven't deeply internalised the civilisational stakes. at Anthropic, the stakes are well understood, but the company believes it has to get there first because nobody else will act responsibly. an Anthropic colleague replied publicly agreeing, putting his own odds above 10% within the decade but the replies are where it gets strange two separate people asked the same honest question. how, specifically, would this kill everyone. neither got a real answer. instead it split into people saying it's just token prediction, people saying we'd be throwing away utopia, and people saying would you rather China wins and one reply that actually engaged with the decision instead of the belief: if you really think this, why resign? stay inside and slow it down one side is arguing about probability. the other is demanding a mechanism. neither is answering the other, which is roughly why this conversation hasn't moved in three years first reported by WSJ, who called it one of the first cases of an Anthropic employee leaving over safety fears his thread: @hilbertspaess dawn.com/news/2028573

Now i’m afraid of AI :)

I built a game for @decentraland’s Friendzone Mobile Buildathon where the buttons are your avatar’s emotes. It’s called Sambung, Indonesian for “to continue,” and also the name of a childhood chain game. One player taps an emote pad. Their avatar performs it in the world, everyone sees it, and it becomes part of the chain. Everyone else has to repeat the entire chain from memory. Get it right → you earn the right to add the next link. Three lives. When the chain dies, its length becomes the record to beat. The hard part wasn’t the game. It was this rule: “Empty venues and single-player experiences without a meaningful social component are not eligible.” That’s a pass/fail gate. And yet every entry is judged by one person, walking in alone, on a phone. So the social layer can’t be something that only activates when a second player arrives. My answer: the record performs itself. Walk in alone, and the players who built the best chain this World has ever held stand on the stage as their own avatars. Each performs the exact link they added, in order, before you’re asked to do anything. Their names and emotes come from the record store, not a script. The room has people in it before anyone else shows up. Two things I had to learn the expensive way: 1. AvatarShape actually renders inside a normal scene. The SDK’s own proto comment says it “is only actually used in the global Avatar Scene.” That comment is wrong. And the entire feature depends on it being wrong. 2. expressionTriggerId and triggerEmote() disagree. expressionTriggerId needs a full URN for an avatar. triggerEmote() takes the short ID. Pass the short form and you get a frozen avatar in the rest pose — silently, with no error. The rest is deliberately tiny. 192 triangles for the entire scene: one cylinder, eight boxes, zero models, zero textures. That’s just 0.016% of Decentraland’s mobile limit, and the build fails if that ever stops being true. There’s also just one 13 KB audio clip, played at eight different pitches. So the chain becomes a melody instead of a sequence of colored flashes. That’s the half of Simon that makes it stick. And the footage in this video isn’t a mockup. A test suite in the repo loads the deployed World in a real Decentraland client, watches the replay, and plays a real chain: WAVE → ROBOT → MONEY While the camera was running, it moved the live weekly record from 2 to 3. You can verify the endpoint yourself. 🎮 Play: decentraland.org/jump/?realm… 💻 Source (MIT): github.com/PugarHuda/sambung Built for @decentraland’s Friendzone Mobile Buildathon on @DoraHacks.