←
Ram6,140 posts
Ram avatar

Ram

@ram_krish2000

trial-and-error

◷ Joined October 2021▧ 964 Photos and videos
Ram avatar
Ram@ram_krish2000·5h

Won 200$ from Walrus x Tatum hackathon Try @ ellai-pay.vercel.app

Hackathon results 🏆 Here are the winners of our Smart Agents for Real Life track: 🥇 Antibody by @linoxbt 🥈 Ferry by @AdwaitKeshari 🥉 Ellai by @ram_krish2000 Agents that remember, pay, and protect, on Walrus Memory and Sui, read through Tatum. More on 3 winners among the 80 projects in the comments 👇

Post attachment
Ram avatar
Ram@ram_krish2000·Nov 14

In the ROBO airdrop, wallets on one public farm list took 87.43% of all the tokens claimed. That is what a drop pays for when nobody screens the list. I spent the SERV hackathon building Thuli to screen it first. You give it a claim list. Code reads each wallet's public history and sets a ceiling. > SERV Reasoning from @openservai applies a written rulebook and writes a reason for every wallet it holds back. > Code checks that answer before anything uses it. > Approved wallets are paid from a vault on Base, where Thuli's own key can post a list and nothing else. I replayed the whole ROBO drop, 35,323 wallets, with only data from before the first claim: 99.03% of the farm-list wallets held 1.00% of a control sample from Hop's airdrop held by mistake 0.05 cents of SERV credit per wallet @reppo's judges read 49 of Thuli's decisions and blocked 11. One rule changed because of it. False holds fell from 2.53% to 1.00% and the catch rate stayed at 99.03%. It is all on a public page, misses included. It is live. A test campaign where you sign one message, pay no gas and get a decision with a reason. A free wallet check. A paid API at $0.05 a wallet over x402, which has already taken a real payment on Base. Try it: thuli-drop.vercel.app Code and threat model: github.com/ramakrishnanhulk2… Built for SERV Hackathon Edition 02, Reppo track. Demo video below.

1:22
Ram avatar
Ram@ram_krish2000·Nov 14

Doji Wave 2 is in for the @MezoNetwork Buildathon. Fixed-rate MUSD loans against locked Bitcoin, now with: - idle lender money earning in sMUSD - rewards that pay your loan down - an early exit for lenders Free test kit, gas paid, just a wallet: dojis.site/quests

1st place in Wave 1 of the @MezoNetwork Buildathon. Doji: fixed-rate MUSD loans against veBTC, while your Bitcoin stays locked and earning. Top grant of 135 projects, 10/10 for Mezo integration. Try it on testnet: dojis.site On to Wave 2. Thanks @akindo_io

Post attachment
Ram avatar
Ram@ram_krish2000·Nov 13

Would you sign this? "Authenticate DAO for confidential intents" a9f14510bca0d490ddf425fcf51c3a599096f8fa598f1bcb0eadd9bbb9bbd50f That is all an approver sees when a NEAR treasury is asked to sign a hash. On my test treasury, the words behind that hash add a new key with full control of the treasury's money on NEAR Intents. It is not rare. In the last year NEAR treasuries were asked to sign a hash 2,622 times, and approved 2,271 of them. I built Kaap at NEAR HACKATHON so members stop signing blind. How it works: 1. A reader runs inside sealed hardware on @out_layer. It reads every new request and works out what it will really do. 2. Plain code compares that with the description. A model on NEAR AI gives a second opinion. It can raise the alarm and never lower it. 3. A guard contract sits in the treasury as a member whose role can vote reject and nothing else. Sputnik DAO allows exactly that. It is not a new multisig: the members vote it in, and can vote it out. On testnet, a request that said "salary for alice" and paid "a1ice" was rejected 25.5 seconds after it was filed. Every step is a transaction you can open. On 298 test requests the guard would stop 41 of 54 thefts and 0 of 244 honest payments. The 13 it misses are published too. The guard can stop a theft. It cannot approve, pay or change a rule, so it can never make one. File a fake theft yourself, no wallet, and watch the guard reject it, usually in under a minute at kaap-snowy.vercel.app/check Code, and one command that checks the proof with no keys at github.com/ramakrishnanhulk2… @NEARProtocol #NEARHackathon

1:51
Ram avatar
Ram@ram_krish2000·Nov 12

My project Moolam won the @chainlink CRE bounty at Monad Metropolis. Chainlink’s network checks every new picture on its own: it fetches it, measures it again and writes the verdict on chain. 47 checks on mainnet since 2 October, all 47 matched, every new one in under 30 seconds. Thanks to the Chainlink and Monad teams for running Metropolis.

Post attachment
Monad@monadNov 11

Monad @monad Nov 11 Article Meet the Metropolis Winners Metropolis is Monad’s global hackathon with over $250k in prizes. 885 teams and 1,256 builders from 76 countries shipped projects on Monad across four tracks: Onchain Finance, Consumer Products,

Ram avatar
Ram@ram_krish2000·Nov 12

Small shops lose their best regulars without noticing. Nobody complains. They just stop ordering, and the owner finds out months later. For the PayPal AI Hackathon (@paypaldev , @devpost ) I built Mend Mend reads a shop's own order history and finds the regulars who went quiet. > Claude turns the owner's one-line goal into a capped plan and writes each customer a short thank-you note from what they used to buy. > The owner approves the exact plan, and PayPal Payouts sends the cashback into the PayPal wallet the customer already pays with. > Only customers who opted in with Log in with PayPal get paid. Every campaign holds back a random group, so the shop sees whether the money brought anyone back instead of guessing. On two years of a UK gift wholesaler's orders, the customers the forecast called likely gone came back on their own 10% of the time. The ones likely to come back anyway did 60% of the time, with no cashback at all. The middle group came back 38% of the time. That middle group is where a nudge can change something, so that's who Mend pays. Claude never picks who gets paid and never sets an amount. Code does the money math, and the owner approves every send. You can try to break it yourself: the demo's Break it step fires real attacks at the live app, like asking Claude to pay $500 a customer or forging a PayPal webhook. Each one is refused. Sandbox only, no sign-up: Demo: mend-web.onrender.com/demo Code: github.com/ramakrishnanhulk2…

2:54
Ram avatar
Ram@ram_krish2000·Nov 11

I asked myself a simple question: can my Hermes agent be tricked into leaking a secret? Nobody could tell me, so I built Othigai, a fire drill for Hermes agents. It's a plugin. > Type /drill and it starts test agents on your own machine, with your model and your own SOUL.md, and plants a trap in an email they read. > Everything they can reach is fake: a fake key, a throwaway folder, a listener on your machine playing the attacker. What I found on free Nous Portal models, same trap, 20 runs each: Beam fell 20 of 20. Crouton fell 2 of 20. Then I tried fixes, 12 runs each. Taking the shell away lifted Beam from 17 to 92 out of 100. One rule in SOUL.md took LongCat to 100 twice, and did nothing for Beam. No single tip works on every model. So Othigai measures the fix on yours. Built for the Hermes Agent Plugin Hackathon @NousResearch Repo: github.com/ramakrishnanhulk2… Catalog PR: github.com/NousResearch/herm…

1:06
Ram avatar
Ram@ram_krish2000·Nov 11

Doji's quest ledger is open on @MezoNetwork testnet. You can try all of it on testnet with just a wallet, on a laptop or a phone. Connect a wallet, press Claim, and free test BTC and MUSD land with no gas. Then borrow, lend and work up from Decchi to Sobashi. EXP has no cash value. I want to see what breaks. dojis.site/quests

Post attachment

1st place in Wave 1 of the @MezoNetwork Buildathon. Doji: fixed-rate MUSD loans against veBTC, while your Bitcoin stays locked and earning. Top grant of 135 projects, 10/10 for Mezo integration. Try it on testnet: dojis.site On to Wave 2. Thanks @akindo_io

Post attachment
Ram avatar
Ram@ram_krish2000·Nov 11

Won 500 USDC, one of 8 @arc Microgrants from @circle out of 371 entries. Within two days of launch, a wallet I don't control started running my demo payroll for the tip. It ran 72 of the first 80 paydays. So far, 327 of 335 paydays were run by wallets I do not control. It only works on Arc, where fees are paid in USDC too, so a stranger pays the fee and earns the tip in the same money. Every contract, test and screen was built in Arc Studio, Circle's coding agent, from my prompts. Thank you to the Circle and Arc team. Try it: ooyaa.vercel.app Code: lnkd.in/g_CvEf6U327 of 335 paydays were run by wallets we do not control. ]

Post attachmentPost attachment

Payday should not depend on the employer being online. I built Ooyaa, a USDC payroll on @arc >Anyone can press Run payday and earn a 0.01 USDC tip. >Each payee is paid once per period, in order, and the contract does the rest. It already works that way. Within two days of going live, a wallet I don't control started running my demo payroll for the tip. It has run 72 of its 80 paydays. The other 7 were me testing. - This only works because USDC is the gas token on Arc. - A stranger pays the fee in USDC and gets the tip in USDC. A proof page re-checks every cent live from the chain. Eight checks, stamped VERIFIED. Arc Studio wrote and tested every contract and screen. Live on Arc mainnet. Not audited, so use small amounts. Built completely using Arc Studio Try it: ooyaa.vercel.app Code: github.com/ramakrishnanhulk2…

1:38
Ram avatar
Ram@ram_krish2000·Nov 10

My AI agent kept running into sellers that want a few cents before they answer. I could approve every payment myself, or hand the agent a wallet and hope. I built a third option for Total Recall with @tatum_io and @WalrusProtocol It’s called Ellai, Tamil for limit, and it runs on Sui Mainnet with real USDC. How it works: My USDC stays in my own wallet, behind a Sui allowance I set: a total cap, a daily limit and an end date. I can end it with one signature. Every time my agent wants to pay, Ellai does one of three things: -> Asks me, when the seller is new or the request is unlike anything I’ve approved. -> Pays on its own, when I’ve said yes to that seller enough times, for close enough reasons, inside my limits. -> Refuses, when the payment is over my limits or I’ve said no to that seller before. A no always wins. My answers don’t sit in Ellai’s database. Each yes or no becomes a memory in my own Walrus Memory account. Ellai searches it by meaning before every payment, and I can read the same memories from Claude Code. The rules are fixed code. No AI model decides about money. Every payment is read back from Sui through Tatum before Ellai counts it as paid. The learning, tested on Mainnet: I labelled 24 requests yes or no in advance. Ellai asked me 8 questions in the first 12, none in the next 12, and let 0 bad payments through. Something I found along the way: a Walrus Memory search could hang forever inside the SDK. I filed it with a script that reproduces it, and the Walrus team confirmed it and opened a fix. Until that ships, Ellai caps every search at 15 seconds, and a search that runs out never pays. What it doesn’t do yet: it speaks x402 version 1 only, a refusal can’t be taken back yet, and nobody outside has audited it. Look around, no wallet needed: ellai-pay.vercel.app Code: github.com/ramakrishnanhulk2…

2:38
Chris Gold 🪙🥇 avatar
⟳ Ram retweeted
Chris Gold 🪙🥇@ChrisGold__·Nov 9

The Nimbles Showdown is live 🏆 Want to earn NIM? Every step pays: 💰 Finish your first matches, earn NIM ⚔️ Duel friends across 18 games, earn NIM 🤖 Beat Nimo on Hard, earn NIM 💌 Invite friends who play, earn NIM 📈 Top the weekly leaderboard, earn big Half goes straight to your wallet, half grows in your Savings Jar Play in Nimiq Pay before Nov 23 👇 nimbles-mainnet.fly.dev @nimiq @NimblesX

0:56

Most "play to earn" games are really luck dressed up as skill. So I built the opposite. Meet Nimbles: 18 skill-only games that live inside @nimiq Pay, where the better player wins every time. There are no dice, no random draws and no loot boxes. 🎯 The games Chess Blitz, 8-Ball Pool, Sudoku, Word Rush, AI Trivia Duel, Solitaire Duel, Spider Solitaire, Bubble Shooter, Last Bite, Darts, Block Puzzle, Speed Math, Domino Chain, 21 Blitz, Bowling, Carrom, Ghost Race and Splat Royale, a 2 to 6 player battle royale. Every match comes down to aim, timing and fast thinking. 🕹️ Four ways to play • Play: free, just for fun • Win: stake NIM against a real rival, and the winner takes the pot • Save: your winnings go straight into your savings jar • Settle: invite a friend and agree on the terms up front, either winner-takes-all or a split you both sign off on 🫙 The savings jar Lock your NIM for 7, 30 or 90 days. Savers earn from an interest pool funded by a 2% fee on Win and Save pots. You can unlock early for a small penalty. 🎁 Fair rewards Chest contents are fixed and earned through play, never random. XP boosts what you earn as a saver. 🔐 Built for real money Sign in with your wallet signature, stakes are held in escrow, and winners are paid out automatically on chain. Nimo, your little host, walks you through it all in the video 👇 Live on mainnet now: nimbles-mainnet.fly.dev Built for the Nimiq Mini Apps Competition.

3:36
Ram avatar
Ram@ram_krish2000·Nov 1

Payday should not depend on the employer being online. I built Ooyaa, a USDC payroll on @arc >Anyone can press Run payday and earn a 0.01 USDC tip. >Each payee is paid once per period, in order, and the contract does the rest. It already works that way. Within two days of going live, a wallet I don't control started running my demo payroll for the tip. It has run 72 of its 80 paydays. The other 7 were me testing. - This only works because USDC is the gas token on Arc. - A stranger pays the fee in USDC and gets the tip in USDC. A proof page re-checks every cent live from the chain. Eight checks, stamped VERIFIED. Arc Studio wrote and tested every contract and screen. Live on Arc mainnet. Not audited, so use small amounts. Built completely using Arc Studio Try it: ooyaa.vercel.app Code: github.com/ramakrishnanhulk2…

1:38